Recent bombings in Iran and Lebanon have displaced hundreds of thousands of civilians. These are not isolated cases: the number of conflict-induced internally-displaced persons (IDPs) has almost doubled between 2014 and 2024, from 38 million in 2014 to 73.5 million in 2024, according to the Internal Displacement Monitoring Center. Our work shows that even temporary displacements may redefine identity boundaries and thereby impact post-conflict political behavior. Hence, while those displaced may return to their homelands, a new divide may overshadow long-term prospects for nation re-building and durable peace.

Intergroup contact theory suggests that hosting IDPs differs in important ways from hosting foreign refugees. In the former case, value convergence between locals and guests is more likely due to closeness in their social status and common inter-group goals, among other factors (Allport, 1954). Yet, exposure to a domestic out-group may still alter the salience of demographic attributes which shape inter-group boundaries, triggering an identity response among locals (Bonomi et al., 2021; Fouka and Tabellini, 2022).

Evacuations in WWII France

Estimating the causal effect of IDPs is challenging because, as with foreign refugees, their settlement patterns are endogenous to local characteristics. In addition, displacement may lead to permanent resettlement, so observed changes in behavior could be the product of demographic and economic re-composition, rather than hosts’ responses to out-group exposure.

Our paper addresses these concerns by leveraging the characteristics of the WWII French temporary civilian evacuation for the French-German border zone. At the onset of the war, French authorities implemented a large-scale evacuation plan targeting the border population. Among the evacuees, Alsatians from the department of Bas-Rhin were sent to the south-west department of Haute-Vienne, more than 700 kilometers away from their homes. According to the Haute-Vienne reception plan, refugees would be spread uniformly across the department (see Panel A of Figure 1). In practice, however, the massive and unexpected influx of evacuees from the border “front zone” – the area between the Maginot line and the border – disrupted the plan. This halted the evacuation of the “rear zone”, a strip of land immediately to the rear of the Maginot line.

Consequently, host municipalities matched with the Alsatian front zone in the reception plan received significantly more evacuees than those matched with the rear: front-zone assignment increased the share of evacuees among locals by 22.8 percentage points. Our identification strategy exploits this variation (see Panel B of Figure 1). It uses the origin zone to which host municipalities were matched as an instrumental variable for the share of evacuees hosted.

What was the hosting experience like for Haute-Viennois? Locals had to share public spaces and, in many cases, their own homes with evacuees. Given the department’s relative poverty, the sudden increase in population placed considerable strain on local resources. At the same time, cultural differences meant that evacuees had to maintain separate institutions, including their own schools and churches. Alsatians were French citizens but had their own regional identity. Inter-group communication was hindered as their knowledge of French was limited: they spoke a dialect close to German, a fact that earned them the nickname “ya-ya” among locals (Bernussou, 2014). Moreover, whereas most of southwestern France was relatively secular, evacuees were, for the most part, devout Christians and more likely to support right-of-center parties. Furthermore, Alsace had been under German rule between 1870 and 1918. Still, evacuees and hosts co-existed peacefully until the former returned to their land a few months after the June 1940 Armistice. Only a negligible fraction of Alsatians permanently settled in the French interior.

Consequences of Hosting Co-citizen Out-groups

We show that temporarily hosting IDPs had important political consequences: municipalities that hosted more Alsatian evacuees experienced an increase in support for left-of-center parties, the dominant political group in the area, after the war. The peak estimates in the late 1940s imply that increasing the share of evacuees by 10 percentage points increased the left-of-center vote share by around 2 percentage points. Difference-in-differences estimates show that while there are no differential trends in prewar vote shares, there is a shift towards left-wing parties in areas matched to the front zone (see Figure 2).

What could have produced this shift in political behavior? Based on the literature, we explore three main mechanisms. First, while the majority of Alsatians were right-wing and locals were predominantly left-wing, some left-wing evacuees may have transmitted left-wing values to their hosts. Second, hosting refugees exposed locals to the costs of war and foreign aggression, consequently influencing political behavior. Third, inter-group contact can change the salience of demographic attributes and redefine inter-group boundaries (Fouka and Tabellini, 2022; Fouka and Serlin, 2026). In particular, exposure to a co-citizen out-group may have strengthened local identity among hosts at the expense of a national one. This identity response may then have translated into shifts in political behavior. In our context, this means that the greater the socioeconomic distance between Alsatian evacuees and their hosts, the more salient the difference between the two groups. As this salience increases, the more likely the political minority among hosts is to converge towards the majority’s behavior and to move away from that of the out-group of evacuees.

An analysis of the treatment effect provides evidence supporting this last channel: Figure 3 shows that hosting evacuees increased left-wing vote share when internally-displaced Alsatians differed most from their hosts. This was the case when IDPs were more likely to be Protestant (Panel A of Figure 3), or when urban (rural) Alsatians were hosted by rural (urban) hosts (Panel B of Figure 3). We also find a greater shift to the left where pre-war host left-wing support was stronger. Hence, it appears that the local pre-war right-wing minority changed their political behavior to better-fit the left-wing majority following an identity response to Alsatian exposure.

Additional results support this explanation. The presence of evacuees led to an increase in post- war civic organizations focused on local recreational activities and the defense of local interests. Similarly, the negative effect on the vote share of right-of-center parties is largely driven by a decline in support for the MRP (Mouvement Re´publicain Populaire), a center-right party with key leaders of Alsatian descent. Finally, the presence of other French refugees, culturally closer to the hosts, did not produce the same effects.

Our results highlight the lasting consequences of temporary exposure to IDPs. These impacts are another hurdle countries must overcome post-conflict. Internal displacements are not just a by-product of conflicts, but another potential mechanism through which conflict begets conflict (Bohnet et al., 2018). IDPs may shape political behavior in the long run as hosts fall back on local identities. These reactions may come at the expense of national unity and influence the success of post-war reconstruction and nation rebuilding.

Jean Lacroix is an Associate Professor in Economics at Université Paris Saclay and a Research Affiliate at CESifo. His research lies at the intersection of Political Economy and Economic History and calls on methods from Economics, Political Sciences, and History. He is particularly interested in post-conflict societies, political transitions and elite persistence.

Ricardo Pique is a Visiting Assistant Professor at Texas A&M University, Mays Business School. His research focuses on the Political Economy of Development and Economic History.

When do people identify with their class? A long line of thinking from Marx and Engels to Acemoglu and Robinson assumes the answer is “always.” Nowhere was this more apparent than 19th century Europe. Elite thinkers on both the left and right believed class politics was inevitable. Lord Salisbury, the British conservative leader, thought that franchise extension in the UK was transforming politics into “a struggle between those who have, to keep what they have got, and those who have not, to get it” (quoted in Ziblatt, p 54). But these predictions turned out to be, at best, mixed. Class politics was central to electoral politics in much of twentieth-century Europe, but in a number of cases – especially the US – class politics never quite consolidated.

In a new working paper, we try to understand when people identify with their class, examining the foundational case of 19th century Britain. Our starting point is that there are many groups that people can identify with, and it is not obvious that their class is the group they will pick.

A large body of work in social psychology tries to understand when people identify with a particular group. Two of the core predictions are that people are more likely to identify with a group if it is higher in status, but they are less likely to identify with it if they are more different from its members. These insights have been incorporated into political economy by Moses Shayo.

Applying this framework to class, the prediction would be that people are less likely to identify with their class when they are culturally different from other members of the same class. If class groups are more culturally diverse, class identity should be weaker. Migration which increases the diversity of class groups should push against the formation of class politics.

The problem with testing this theory is that class identity is hard to measure. If we want to study whether changing cultural distance to the working class affects whether people identify with the working class, we would want a measure of whether they identify with the working class. Unfortunately, lots of things that do partly proxy for class identity, like the neighborhoods people live in or the schools they send their children to, are also proxies for income.

Children’s names give a relatively clean measure of class identity. The name you give your child is a free choice, both in that it is your choice, and in that it doesn’t cost anything. Class background is very predictive of the names people give their children. Accounting for factors like occupation and income, a more working class name should be more reflective of identification with the working class.

Using data on all people in the 1881 census of England and Wales, we score each name by how distinctive it is to people with occupations that were considered working class at the time, relative to those considered upper or middle class. The most working class names are Patsy, Pat, Mickel, and Bridget. The least working class are Cyril, Reginald, Cecil, and Gerald. These are the kinds of names that Oscar Wilde gave upper class characters. Calculating the scores of MPs, we see that Labour MPs and MPs affiliated with trade unions have higher scores, while aristocrats, alumni of elite schools and universities, and members of social clubs have lower scores.

The names people gave their children also reflected their class positioning. Figure 1 plots the average score for children recorded in the 1911 census against the status of their father’s occupation (using scores from the HISCAM). In general, fathers in high status occupations gave their children less working class names. But the observations off the regression line are interesting. Clerks were not particularly high status, but did not work with their hands and were considered lower middle class. They tended to give their children less working class names than other occupations of the same status.

In the census, we observe when children were born and the parish in which they were born, in addition to their names, which we use to measure class identity. Combined, this data gives us a time-stamped record of each family’s class identity and location in the year of each birth. For families with multiple children, we have multiple measures of class identity from points in time and different locations. If a family moved to a location where it was more culturally distant from the working class, we would expect that family to identify less with the working class, as measured in the names of subsequent children. Similarly, if a family lived in a location where in-migration increased the cultural distance of the average working class resident from that family, we would expect that family to identify less with the working class. Our research design effectively compares the changes in outcomes in these kinds of families, to the changes in outcomes in families that stayed in places that did not experience much in-migration.

To measure cultural distance, we construct a measure of the cultural similarity of different parishes in England and Wales for the working and upper classes, based on marriage records. The idea is that if two parishes are culturally similar, we should see more marriage between people born in the parishes. With this measure of cultural distance, and data from the census on the inhabitants and occupations of every parish and where they were born, we can measure the cultural distance between each person and the average upper or working class resident of their parish. Our focus is on average distance to the parish’s working class residents, relative to its upper class residents. We then look at how changes in this measure of distance affect class identity as measured through names.

In the top row of Figure 2, we look at the relationship between the change in working class scores between two children in a household, and the change in the family’s relative distance to the working class in the birthplaces of the children, relative to the upper class. Increasing distance to the working class corresponds to a shift away from working class identity. Going from the 5th to the 95th percentile of the relative class distance distribution corresponds to a decrease in working class identity of around 11% of the difference between Oxbridge- and non-university-educated MPs. If we separate out families where the two children were born in different locations (”movers”) and those where both children were born in the same location (”stayers”) we get similar estimates. This is reassuring in that it suggests that the mechanism we want to study, and not something else that could be causing households to migrate, is driving our estimates.

To interpret this relationship as causal, we would need to believe that in the absence of the change in relative class distance, the households that experienced chages in relative class distance would have followed similar trajectories in class identity to those that did not. In the second row we look at changes in the class content of the names of the previous two children. We don’t find any evidence that households were already shifting away from the working class before they moved to locations with greater cultural distance from the working class. This helps to validate the assumption, suggesting these different groups of households were following similar trajectories before the move. We also don’t find any evidence that these different groups gave their children more or less working class names, had more or less working class names, or had more or less working class occupations before the changes in class distance due to migration.

Did these changes in class identity matter for politics? If people identify less with the working class, we would expect them to be less supportive of political movements that promote the wellbeing of working class people. In late 19th century Britain, the main such movement was the Labour Party. If a place attracted lots of culturally-distant working class migrants, that would increase the average cultural distance to the working class in that place, leading people to identify less with the working class and making them less likely to vote for Labour.

In Figure 3 we examine that prediction. The figure shows the relationship between the change in relative class distance at the local level, 1891--1911, and voting for the three main parties in that area in successsive elections. In the 1880s and 1890s, the places that experienced increases in relative class distance were no more likely to vote for one party than any other. But between 1892 and 1910, these places experienced a relative shift away from Labour.

We also find evidence of politicians using rhetoric less targeted at working class voters in these areas. At the individual level, workers who were more culturally distant from the working class were less likely to join labor unions, even when comparing individuals in the same occupation in the same parish.

But what were these areas experiencing rising class distance? The main form of migration in 19th century Britain was from rural to urban areas. Peasants migrated into the cities and became industrial workers. Because of this process, by 1881, when we begin our analysis, cities had the highest cultural distance to the working class. In contrast, agricultural regions had relatively homogeneous working class populations. Over the period we study, however, these patterns reversed. In rural areas, out-migration continued, reducing the share of locally-born adults. But in urban areas working class distance actually decreased, because a growing share of the population was locally-born, the children of previous rural-urban migrants.

In the left panel of the figure below, we plot the (binned average) change in class distance between 1891 and 1911 against the share of employment in agriculture in 1891. Rural areas experienced an increase in relative class distance, but urban aread experienced a decrease. In the right panel we plot the change in voting for Labour against agricultural employment. The Labour Party gained votes in those urban areas which had falling class distance. Industrialization and migration first delayed and then accelerated the formation of urban working class politics in the UK.

Subscribe now

Vicky Fouka is an associate professor of Political Science at Stanford University. She studies political economy and political behavior from a comparative and historical perspective. Her research examines what shapes social identities in the short and long run and their implications for political and economic behavior and policy design. Some topics she has focused on include immigrant integration, prejudice against ethnic and racial minorities, and intergroup conflict.

Theo Serlin is a lecturer in the department of Political Economy at King’s College London. He studies international and comparative political economy. His research integrates economic geography into political economy models of policy preferences and electoral politics. He is especially interested in the politics of trade, migration, and economic change.

The French Revolution was a watershed moment in world history. In just a few years it dismantled a centuries-old political order, recast citizenship and sovereignty, and propelled ideas about rights and representation across Europe. Although the causes were many, taxation has long been singled out as central (Noberg, 1994; Touzery, 2024). As Alexis de Tocqueville put it, allowing monarchs to tax without consent “sowed the seed of practically all the vices and abuses” that undermined the regime (Tocqueville, 1856). Yet, despite the prominence of this idea in historical narrative, systematic evidence linking fiscal burden to popular unrest has been limited.

Our new paper, Extractive Taxation and the French Revolution, tests this hypothesis with newly digitized fiscal and unrest data. We geo-reference bailliage-level measures of per-capita tax burden circa 1780 assembled from Touzery (2024), and map the substantial spatial variation in that burden (Figure 1). We merge those measures with the HiSCoD catalogue of riots assembled by Chambru and Maneuvrier-Hervieu (2024) to ask whether areas bearing heavier fiscal loads experienced more popular revolts in the decades leading up to the French Revolution.

Figure 2, Panel A, plots a simple binscatter and illustrates the basic relationship: bailliages with higher per-capita taxes had more riots between 1750 and 1789. The association is robust to controlling for several factors commonly linked to pre-revolutionary and revolutionary unrest. Moving from a bailliage in the bottom quartile of the tax-burden distribution to one in the top quartile (about an 8% share of per-capita income at the time) more than doubles the number of riots in 1750–1789. We interpret these patterns as evidence of broad opposition to taxation. To probe the mechanism, we turn to the cahiers de doléances (literally, the lists of grievances) compiled ahead of the Estates General in the spring of 1789. Figure 2, Panel B, shows that bailliages with heavier tax burdens submitted more complaints about taxation. Notably, these complaints emphasized not only the economic burden of taxation but also its unequal incidence across social groups and territories and its extractive character.

Historical accounts emphasize that indirect taxes were especially resented (Sands and Higby, 1949; Touzery, 2024). Administered through a dense fiscal infrastructure of warehouses, checkpoints, and inspections run by the Ferme générale, these levies were highly visible in everyday transactions. The salt tax gabelle—given the state monopoly over a basic necessity—and the traites—which fragmented internal trade through tolls and inspections—were widely viewed as particularly oppressive, and their borders often reflected centuries-old bargains rather than economic fundamentals. Consistent with this view, indirect taxes drive our results: when we decompose the overall tax burden, the indirect component is the strongest predictor of both riots and tax-related complaints.

This motivates a quasi-experimental strategy exploiting sharp discontinuities in the salt tax and the traites, where neighboring municipalities faced markedly different indirect tax regimes. We find that crossing the border from a low- to a high-tax municipality increases the number of riots. The effects begin to emerge in the 1760s, and become larger and more precise over time, peaking in the 1780s. Crossing the tax frontier roughly doubles the number of riots between 1780 and 1789. Consistent with our interpretation, the effect is driven entirely by tax-related riots, while we find no discontinuities for unrest unrelated to taxation, such as food or labor riots.

One interpretation of these findings is that taxation created latent conditions for unrest, which were activated by broader political and economic forces. One such force was the spread of Enlightenment ideas about equality and justice (Darnton, 1982; McMahon, 2001). Consistent with this view, the effect of crossing a tax border is larger where tax gaps are greater and along borders with stronger diffusion of Enlightenment ideas, where fiscal inequality would have been more salient. A second trigger was economic distress driven by adverse weather shocks in the late 1780s (Lefebvre et al., 1947; Waldinger, 2024). Poor harvests raised wheat prices and strained household budgets, especially in heavily taxed areas. Combining the RD design with spatial and temporal variation in temperature shocks, we find that a 10% increase in summer temperature (about 1.8°C) roughly doubled the number of riots in high-tax municipalities relative to their low-tax neighbors.

Local grievances can generate isolated disorder; if they spread, they can fuel revolutionary upheaval. To assess whether fiscal discontent shaped this transition, we study the Grande Peur (Great Fear) of July–early August 1789 (Lefebvre, 1973). Following the fall of the Bastille, rumors of aristocratic plots swept rural France, triggering attacks on manor houses and the destruction of feudal records, and culminating on August 4 with the abolition of feudal privileges. Tracing the spatial diffusion of this panic, Figure 3 shows a clear divergence: high-tax bailliages were reached earlier and experienced faster spread. By August 4, a substantially larger share of high-tax territory had already been affected, and by the end of the episode roughly 80% of high-tax areas had been reached, compared with about 60% of low-tax areas. Our formal analysis confirms that high-tax bailliages were more likely to ignite panic contagion and experienced earlier and more extensive spread overall.

These patterns suggest that fiscal grievances fueled the revolutionary crisis from below. But did taxation also shape the Revolution from above, through the behavior of its political representatives? To answer this question, we analyze more than 60,000 speeches from the Archives Parlementaires, delivered between May 1789—when the États Généraux convened—and January 1793, when Louis XVI was executed. Deputies from high-tax constituencies engaged very differently with fiscal issues than their counterparts from low-tax areas. They were about 60% more likely to speak about taxation, nearly twice as likely to criticize the Ancien Régime, and roughly 60% more likely to defend the Revolutionary project in tax-related speeches. Their rhetoric also framed taxation more often as unequal or oppressive and more frequently called for fundamental fiscal reform.

We then move beyond fiscal debates to examine how taxation shaped political behavior at key turning points of the early Revolution. In the weeks following the Great Fear—when the Assembly was actively debating the future of the regime—deputies from high-tax constituencies were more likely to demand institutional change, explicitly call for the abolition of feudal privileges, and openly criticize the monarchy. Turning next to the Assemblée Législative (October 1791–September 1792), we find that legislators from heavily taxed areas were more likely to support the abolition of the monarchy. Finally, using newly digitized roll-call votes from the Convention Nationale, we show that deputies from high-tax regions were also more likely to vote for the king’s execution in January 1793.[1]

Taken together, our findings suggest that taxation played a central role in the origins and dynamics of the French Revolution. More broadly, the results speak to a mechanism through which fiscal systems that rely on coercion and inequality can destabilize regimes. Future work could examine whether similar dynamics operated in other historical settings—such as Imperial Russia or late-Qing China—where extractive taxation coexisted with social inequality and political transformation, to assess how far these patterns travel across time and institutional contexts.

Subscribe now

REFERENCES

Chambru, C. and Maneuvrier-Hervieu, P. (2024), ‘Introducing hiscod: A new gateway for the study of historical social conflict’, American Political Science Review 118(2), 1084–1091.

Darnton, R. (1982), The Literary Underground of the Old Regime, Harvard University Press, Cambridge, MA.

Degrave, A. (2023), ‘Local rule, elites, and popular grievances: Evidence from ancien régime france’, Journal of Historical Political Economy 3(1), 1–29.

Lefebvre, G. (1973), The Great Fear of 1789: Rural Panic in Revolutionary France, Princeton University Press.

Lefebvre, G., Palmer, R. R. and Tackett, T. (1947), The coming of the French Revolution, Princeton University Press Princeton.

McMahon, D. M. (2001), Enemies of the Enlightenment: The French Counter-Enlightenment and the Making of Modernity, Oxford University Press, Oxford.

Norberg, K. (1994), The french fiscal crisis of 1788 and the financial origins of the revolution of 1789, in ‘Fiscal crises, liberty, and representative government, 1450-1789’, pp. 253–298.

Sands, T. and Higby, C. P. (1949), ‘France and the salt tax’, The Historian 11(2), 145–165.

Shapiro, G., Tackett, T., Dawson, P. and Markoff, J. (1998), Revolutionary Demands: A Content Analysis of the Cahiers de Doléances of 1789, Stanford University Press.

Tocqueville, A. d. (1856), ‘The old regime and the french revolution’, New York (Original: L’Ancien Régime et la Révolution, Paris 1856).

Touzery, M. (2024), Payer pour le Roi. La fiscalité monarchique France, 1302-1792., Champ Vallon, Paris.

Waldinger, M. (2024), ‘“Let them eat cake”: drought, peasant uprisings, and demand for institutional change in the French Revolution’, Journal of Economic Growth 29(1), 41–77.

[1] Political authority during the Revolution was exercised through three successive national assemblies. The Assemblée Constituante (1789–1791) drafted the constitutional framework of the new regime; the Assemblée Législative (October 1791–September 1792) governed under the Constitution of 1791 until the fall of the monarchy; and the Convention Nationale, established after the monarchy’s abolition, ruled France from September 20, 1792, to October 26, 1795.

Tommaso Giommoni is an Assistant Professor at the Amsterdam School of Economics, University of Amsterdam, a Research Affiliate at CESifo, and a Research Fellow at the Tinbergen Institute. He is also affiliated with the Baffi CAREFIN Centre at Bocconi University. His main research interests lie at the intersection of Public Economics, Political Economy, and Economic History, with a particular focus on topics related to taxation, economic growth, and inequality.

Gabriel Loumeau is a full professor of Economics at the University of Neuchâtel, where he holds the Chair of International Economics. He is affiliated with the Center for Economic Policy Research (CEPR) and the CESifo research network. His research lies at the intersection of urban, regional, and international economics. He studies the causes and consequences of regional development, with a particular focus on how economic geography, infrastructure, and public policy shape patterns of growth, commuting, and inequality.

Marco Tabellini is an assistant professor in the Business, Government, and International Economy unit and is affiliated with the National Bureau of Economic Research (NBER) and the Center for Economic Policy Research (CEPR). His research studies how international and internal migration reshape politics, society, and the economy, with a focus on when and why immigration generates political backlash, the conditions under which social integration succeeds, and how migration transforms societal boundaries in countries such as the United States. More broadly, his work in political economy examines how institutional outcomes and democratic processes are shaped by internal forces (such as taxation) and external forces (such as trade).

The French countryside simmered with discontent in the century prior to the French Revolution. As the monarchy sought to assert greater authority and control across its territory, building its informational and infrastructural capacity to levels not previously seen, it was met with grievances and rebellion both far and wide (see the map below). Commoners targeted symbols and agents of the royal authority, protesting tax collection, military conscription, and what they perceived as arbitrary or the unfair pursuit of justice. These rebellions would ultimately help set the stage for the French Revolution, one of the most influential and heavily studied political events of the last 250 years.

The Long and Contested Process of State Formation

The emergence and spread of modern nation-states over the past five centuries has been one of the most consequential developments in human history, driving everything from transformative economic development to powerful bureaucracies, the rise of standardized language, and even world wars. Yet building a strong state is a long and complex process. It has taken most states centuries to overcome powerful fragmented regional and religious elites that had private armies and taxation rights, and many states around the world still struggle with fundamental weaknesses.

Along the way, states have sought to penetrate society and centralize control and administration. However, unlike the ultimate result of state-building—which entails social order through the monopolization of violence by the state, the capacity to enforce rules, and central administration of territory and social groups—the process of state-building is far messier and often riven with contestation and contention. Our article “State-Building and Rebellion in the Run-Up to the French Revolution” demonstrates this in the context of pre-revolutionary France, arguably one of the world’s most precocious state-builder, and one which guided and pushed state-building in its neighborhood in continental Europe.

Centralization and Communication as State-Building

A core component of state-building efforts is the expansion of communication routes. By connecting localities more directly to central authorities, this infrastructure strengthens the ability of the state to penetrate society through the control and administration of local activity and rules. Yet the material consequences of increased—though still incomplete—state power can generate social contestation and popular resistance stemming from the greater local capacity of the state to enforce taxes, carry out rules of justice, and conscript civilians. It can also result from the crowding out of private interests and activities in spaces that the state enters. At the same time, the state may lack sufficient coercive power and legitimacy to deter challenges to its authority. The result is a push and pull between the state and society that periodically devolves into violent civil resistance.

The French Horse-Post System as the Monarchy’s Eyes and Ears Across the Kingdom

France was caught in precisely this dilemma throughout the eighteenth century. A central element of the monarchy’s state-building efforts was the expansion of the horse-post relay network, which nearly doubled in size over the period (see image).

With cars, trains, and telephones still far in the future, these relays established attended lodging quarters and a well-prepared set of fresh horses for messengers carrying information for the royal administration where there had previously been no support. The horse post represented one of the primary means of consolidating the hierarchical French state’s informational capacity as it sought to rule and control a fragmented territory. Located at the entrances to towns or at rural crossroads, horse-post relays were prominent waypoints, comprising stables, lodging quarters, and oftentimes a tavern. Young, sharply dressed horse-post messengers flashed through the countryside between relays (see image). The wealthy local notables that came to staff these waypoints as postmasters effectively doubled as intelligence agents on behalf of the central administration by monitoring passengers and communicating information with regional officials.

Consequences of the Horse Post for Local Order and Rebellion

The development of the horse-post relay network had several consequences. At the national level, it enabled the rapid communication essential to the effective functioning of the state. It allowed central officials, tax farmers, and provincial intendants and local tax officials to coordinate swiftly through directives, reports, and fiscal orders on matters of tax collection and enforcement. That included efforts to suppress salt smuggling across the kingdom’s disparate and consequential salt-tax zones (gabelles). This coordination required frequent and timely exchanges between central administrators and local agents, who were responsible for estimating tax yields and implementing their collection. The horse-post system also facilitated the transmission and enforcement of military recruitment orders, enabling directives to travel efficiently from central authorities to intendants and onward to local recruitment and conscription officers.

Beyond these nationwide implications, the expansion of the horse-post relay network disrupted local social equilibria. Some local actors benefited from these developments. For instance, postmasters gained tax exemptions for their services and other privileges. Local elites also likely benefited from the establishment of horse-post relays, as improved communication with provincial and central authorities could have enhanced their influence and elevated their status.

But others lost out. The various monopolies associated with the horse post crowded out competing private interests. Horse-rental businesses were severely disrupted by the ban on renting their horses along roads connecting horse-post relays. Local farmers and innkeepers situated off postal roads also suffered, as postmasters had priority access to hay and animal fodder for their relays. Additional social groups like commoners, peasants, and merchants may have also lost out as relays brought increased surveillance and stricter enforcement of state regulations. Indeed, enhanced state communication through the horse-post relay network likely made it easier for authorities to monitor and enforce tax collection, military conscription, and judicial decisions.

Backfire During the Process of State-Building

To test this hypothesis, we combine original archival data on the development of the horse-post relay network over the eighteenth century with a comprehensive database of rebellions in pre-Revolutionary France. We digitized one edition of the comprehensive lists of horse-post relay locations (Liste des postes) per decade from the 1710s through the 1790s and mapped them across France. We also digitized comprehensive data from the Jean Nicolas survey of eighteenth-century rebellions, which documents 6,000 rebellions over this period along with their locations, number of participants, and targets. We geolocate these data to roughly 35,000 parishes, the most granular units of administration in Ancien Régime France, and one which no prior study has analyzed during this period.

We then use a staggered difference-in-differences design that compares changes in rebellion in parishes that received (or lost) a horse-post relay to those that would later receive (or lose) one. This strategy accounts for fixed parish-level characteristics that might influence both the likelihood of receiving a relay and the propensity for rebellion.

While the initial spatial configuration of the network in the sixteenth and seventeenth centuries reflected long-standing strategic political and military concerns, its development over the eighteenth century was primarily driven by the state’s efforts to expand and centralize its authority across the interior of the kingdom, resulting in the densification of the network around major regional nodes. Key to our identification, although the placement of central nodes of the relay network could have been endogenous to regional population dynamics, its local configuration between these nodes was plausibly exogenous, as it was designed to minimize travel time along pre-existing roads, subject to the constraint that relays were spaced 10–15 kilometers apart due to the physiological limits of horses. We exploit local variation in the network through decade-by-administrative division fixed effects, principally relying on within-canton variation—cantons gathered on average 7 parishes and 5,000 inhabitants, with a radius of 6 kilometers.

We find that the introduction of a new horse-post relay in a parish was associated with more local rebellions in subsequent decades (see figure below). We attribute this finding mainly to the material consequences of state efforts at penetrating and ordering local society as it centralized and enhanced its informational capacity. Horse-post relays were strongly associated with rebellions against agents with coercive powers to maintain and enforce order: the military, the police, and the judiciary. They were also associated with rebellions against various forms of taxation as well as rebellions by notables that had private interests along postal roads. Furthermore, while these relationships broadly prevail and encompass the lion’s share of rebellions, we also find a disproportionate response of rebellions to horse-post relays in places of pre-existing state fragmentation and territorial divisions. These findings demonstrate an increasingly powerful state that nonetheless struggled to control local activity across a heterogeneous territory as many of its efforts were contested.

Our empirical approach and additional data give us a unique opportunity to evaluate and challenge several alternative explanations, namely the erosion of traditional social hierarchies, the role of war zones and war involvement, information and collective action possibly spurred by the contemporaneous letter-post system, changes in the recording of rebellions by police brigades charged with keeping order, and the contemporaneous transit infrastructure.

Implications

Our findings have implications for the scholarly understanding of the co-evolution of states, order, and disorder, as well as for the vast scholarship on the origins of the French Revolution. Most importantly, they underscore the importance of conceptually distinguishing the process of state-building from state strength itself. While greater state capacity may ultimately support political stability and order, the process of state-building itself can be disruptive to pre-existing social structures and contested, even for decades at a time. This process likely fueled the accumulation of grievances and repertoires of resistance that ultimately exploded during the Revolution. Indeed, exploratory analyses suggest that, while the horse-post relay network in itself did not facilitate information diffusion among citizens or reduce their collective action costs, parishes that received relays—and experienced an associated uptick in rebellion—were more likely to later host political societies that were critical players during the Revolution.

Michael Albertus is a Professor of Political Science at the University of Chicago. He studies how countries allocate opportunity and well-being among their citizens and the consequences this has for society, why some countries are democratic and others aren't, and why some societies fall into civil conflict. His newest book, Land Power: Who Has It, Who Doesn't, and How That Determines the Fate of Societies (Basic Books, 2025), examines how land became power, how it shapes power, and how who holds that power determines the fundamental social problems that societies grapple with.

Victor Gay is an assistant professor of economics at the Toulouse School of Economics. His work is at the intersection of economic history, political economy, and labor economics.

Not long ago, history and machine learning seemed like fundamentally different worlds. Today, these worlds are merging in ways that would have seemed quite unlikely until recently, and thus are even more exciting. The release of ChatGPT almost three years ago marked a turning point. What initially introduced LLMs to a broad audience quickly proved to be relevant for the academic community as well, spreading rapidly across disciplines in both the natural and social sciences, including history.

With this in mind, we organised a Workshop on History LLMs at the University of Zurich and ETH Zurich in early December. The workshop brought together a wide range of researchers from history, economics, and computer science (CS) to reflect on the current state of the field and to discuss where it may be heading.

The more than 40 applications we received highlighted the strong and ongoing interest in the topic and its resonance within the field, reinforcing our view that the workshop content closely mirrored the broader intellectual landscape.

Subscribe now

The submissions and discussions revolved around three key themes:

1. Using LLMs to measure and quantify the past

2. Conceptualising History LLMs, what is a History LLM and what makes them distinct

3. Implementing History LLMs in practice

1. Measuring and Quantifying the Past with LLMs

One of the most immediately useful and widely adopted uses of LLMs in economic history research is as highly capable research assistants. LLMs can process vast quantities of text at scale. They do so consistently and with few restrictions on language. Using “helpers” at such a scale would be infeasible or unaffordable using human labor.

Historians and social scientists use LLMs to:

• Extract information from large text corpora

• Generate summaries or classifications

• Explore potential analytical avenues
  

In this sense, LLMs augment human labour. They do not replace researchers, but they dramatically expand what an individual researcher or small team can accomplish. The early narrative around productivity gains might have been overly optimistic, sometimes suggesting that LLMs could function as fully autonomous research assistants or even digital clones of researchers. This is arguably exaggerated. However, we are already living in a period where scholarly manpower is significantly amplified by LLM-based tools.

Recent work presented at the workshop showed how LLMs are used to transform unstructured historical sources into quantitative evidence.

1. Koschnick shows how transformer-based language models can operationalise innovation and knowledge spillovers by measuring semantic novelty and cross-field similarity in publication titles and patents from 1600–1800, directly testing Mokyr’s theory of a feedback loop between propositional and prescriptive knowledge.

2. Chen, Duffy, and Weigand use contextualised transformer embeddings to recover and quantify distinct historical meanings of “race” in British scientific writing, tracking shifts between biological, analytical, and social conceptions over the nineteenth and early twentieth centuries without imposing modern categorisations.

3. Ash and Xue apply LLMs to interpret and classify historical proverbs, enabling the measurement of cultural values such as work ethic, individualism, and innovation across regions and over time during the British industrialisation.

4. Cummins and Noble use LLM-assisted pipelines to extract religious language and economic information from United Kingdom wills spanning 1300–1850, quantifying long-run changes in sincere versus nominal religiosity.

Together, these studies show how LLMs function as scalable tools for measuring beliefs, culture, and knowledge production in the past. The sheer scale of applications, categorizing complex concepts quickly and at scale, dramatically expands what researchers can accomplish using text.

For historical research, this means that phenomena once considered too large, too dispersed, too complex, or too text-heavy can now be studied systematically. Quantification, pattern detection, and large-scale text analysis are becoming standard components of researchers’ toolkits.

2. Conceptualising History LLMs

While commercial LLMs are powerful, they face a fundamental limitation for historical research: they cannot forget. Once trained, the weights of a model are fixed. While they can be retrained using fine-tuning and similar techniques, LLMs inevitably encode knowledge of later events because they are trained on modern corpora. No prompt, nor even the most advanced prompt engineering, can fully remove this hindsight bias. In the language of CS, modern-trained LLMs are “contaminated” with a knowledge of what is to come later.^[1]

A similar problem is faced by historians themselves. A historian studying the Munich Agreement in 1938 cannot un-know the Holocaust and World War II, and neither can a modern LLM. This “curse of knowledge” fundamentally limits our ability to reconstruct how past actors understood uncertain futures.

1. Varnum et al., however, point toward an exciting solution. They argue that LLMs can become powerful tools for historical social science if they are carefully constrained. History LLMs can be understood as language models trained on historical corpora that encode the cultural, psychological, and normative patterns of past societies. Crucially, their distinctiveness lies in temporal specificity: a true History LLM reflects the mental worlds embedded in historical texts only up to a clearly defined cutoff point and therefore differs systematically from modern LLMs trained on contemporary data.

2. Ahmad highlights important caveats that accompany the training of History LLMs by situating them within the political economy of knowledge. He argues that History LLMs are shaped by asymmetries in archival survival, digitisation, language dominance, and AI infrastructure. As a result, they are never neutral reconstructions of the past. Rather, they constitute algorithmic condensations of historically contingent archives and therefore carry the risk of reproducing colonial, geopolitical, and epistemic hierarchies.

For historians, such History LLMs resonate strongly with long-standing methodological traditions. Collingwood emphasised the importance of reimagining past thought, while economic historians such as Robert Fogel used counterfactual reasoning to understand historical causality. LLMs offer a novel way to operationalise these approaches. Unlike traditional models with rigid assumptions, LLMs learn complex patterns of reasoning, language, and cultural nuance directly from text.

Projects such as MonadGPT and StoriesLM represent important early steps toward using LLMs to model historical reasoning. However, in our view, these efforts do not yet come close to realising the full potential of LLMs for historical knowledge production and counterfactual reconstruction.

3. Implementing History LLMs

Implementing History LLMs is both technically and conceptually demanding. Doing so requires several key ingredients:

• Carefully curated and temporally bounded training data

• Explicit enforcement of historical knowledge cutoffs

• Systematic evaluation of historical accuracy, recall, and anachronism

While discussion around such models has circulated for some time and again gained considerable momentum recently^[2], research presented at the workshop demonstrates that these ideas are rapidly moving from conceptual proposals toward concrete implementations.

Researchers from the University of Mannheim and our team in Zurich presented approaches to training History LLMs with the aim of reconstructing historical belief spaces, forecasting elections, and analysing political or cultural reasoning without contamination from later events.

1. One approach, presented by Cheng, Kranz, and Ahnert, combines curated historical corpora with prompting strategies that assign models a temporal identity. This is supported by validation layers designed to detect anachronistic information and enforce cutoff years. While this strategy goes beyond naïve prompting, it still relies on post hoc controls rather than fully eliminating future knowledge from the model. As a result, the next step the team aims to take is to train their own base model on time-restricted historical data.

2. A hurdle already tackled in the work of Göttlich et al. through the training of their own history LLM, Ranke-4B^[3]. Rather than relying primarily on prompting strategies, they pretrain and fine-tune language models exclusively on texts available prior to specific historical cutoff dates. These time-locked models more reliably reproduce period-specific beliefs, uncertainties, and normative judgments, thereby reducing anachronistic contamination and better approximating the informational horizons faced by historical actors.

Detailed results and training specifications for the University of Zurich model-family, Ranke-4B, are documented in our public GitHub repository, which will soon outline the full training pipeline, validation strategy, and model design choices. Additional updates and results will be added gradually in the near future.

Across approaches, effective implementation hinges on careful benchmark design, rigorous validation against historical knowledge constraints, and a balance between computational feasibility with historical fidelity.

When trained exclusively on temporally bounded corpora, History LLMs offer something genuinely new: scalable and systematic tools for reconstructing historical “thought worlds.” They allow researchers to explore how people in the past might have reasoned under uncertainty, without importing modern hindsight into historical analysis.

Looking Ahead

In 1985, Steve Jobs speculated that one day software might allow us to “ask Aristotle a question and get an answer.” What once sounded like science fiction now seems within reach.

History LLMs are not replacements for historians. However, they represent a new set of tools and instruments that can deepen our understanding of the past, allowing us to scale long-standing historical research in new ways. They also hold out the promise of extending measures of attitudes and beliefs backwards in time, in a rigorous and quantifiable way, through the training of time-locked LLMs.

The Zurich workshop made one thing clear: the field is moving fast, but the most exciting work lies ahead. Measuring the past, conceptualising historical reasoning, and implementing time-locked models are all steps toward a new kind of historical inquiry.

^[1] See e.g. Jens Ludwig, Sendhil Mullainathan, and Ashesh Rambachan, “Large Language Models: An Applied Econometric Framework,” NBER Working Paper 33344 (2025), https://doi.org/10.3386/w33344.

^[2] See e.g. recent X contributions by @Teknium, @victormustar and @JonhernandezIA.

^[3] Leopold von Ranke: „wie es eigentlich gewesen“ [how it actually was]

Daniel Göttlich is a Predoctoral Fellow at the University of Zurich with a particular interest in applying machine learning and language modelling to historical data. His research lies at the intersection of political economy, economic history, and cultural economics.

Dominik Loibner is a Research Fellow at the University of Zurich, working at the intersection of cultural economics, political economy, and economic history, with a focus on large-scale historical data and the application of machine learning techniques.

Hans-Joachim Voth is UBS Foundation Professor of Economics at the University of Zurich and Scientific Director of the UBS Center for Economics in Society. His research spans economic history, political economy, and cultural economics, with growing applications of computational social science. He has contributed extensively to research on Europe’s long-run economic development. His work also examines cultural persistence and change, including national identity, social networks, and extremist ideology, and more recently applies vision AI to large-scale U.S. yearbook data to study long-run trends in individualism.

Across Europe, the first half of the 20^th century brought civil wars, fascist takeovers, foreign occupations, genocidal violence, and entrenched authoritarian rule. The effects of this violence and repression outlasted the end of regimes, often being transmitted across generations. They left lasting marks on voting behaviour, party systems, and attitudes towards violence and justice that are still visible today in electoral maps, views on democracy and minorities, and in how communities remember (or minimize) the past.

The link from past to present is not straightforward. The same civil war or authoritarian regime can set some communities against violence and leave others more disposed to it, and exposure to repression can fuel demands for accountability in one context but stir nostalgia for “order” in another.

A new Comparative Political Studies special issue on Legacies of Repression and Resistance in Early Twentieth-Century Europe explores this variation to ask when episodes of mass violence, mobilization, and authoritarian control reproduce themselves, and when they instead generate backlash, new cleavages or efforts at redress. Authoritarian rule rests not only on constitutions and regime labels, but on local practices of violence, control, co-optation, and mobilization that often outlive the dictatorship itself.

A Typology of Legacies

In this approach, legacies are not vague echoes but causal links between a historical episode and a later outcome, carried by identifiable mechanisms. A legacy exists when a contemporary pattern cannot be understood by present-day causes alone because it still bears the imprint of an earlier conflict or institution.

The special issue introduction distinguishes two broad families. On the one hand, in legacy congruence, the phenomenon observed today resembles the one in the past: past violence predicts later violence, partisan loyalties resurface under new party labels, repertoires of mobilization survive regime change and return when institutions weaken. On the other hand, in legacy alteration, the phenomenon observed today is of a different kind compared to the past political phenomenon: violent conflict leaves behind lasting electoral divides rather than renewed war; armed resistance is recast as peaceful civic campaigns; authoritarian institutions evolve into semi-democratic arrangements that retain informal hierarchies. This typology – contrasting congruent and altered legacies – sharpens what is meant by a “legacy” and shifts attention from showing that history matters to explaining how it continues to do so.

Moreover, the introduction identifies seven legacy types—persistency, discontinuity, creation, reactivation, transformation, substitution, and blunting—and emphasizes that legacies must be understood as causal chains linking past phenomena to present-day outcomes, each mediated by specific mechanisms.

Legacies in Practice: Evidence from Five Countries

The country studies put this typology to work on concrete histories, showing how similar kinds of repression and resistance can leave very different marks.

Finland: class conflict and the ballot box

In Finland, the key episode is the 1918 Civil War between insurgent “Reds” and government “Whites”. Municipalities that saw more lethal violence did not relapse into conflict. Instead, they developed distinctive electoral profiles for decades: support for left-wing parties declined, while conservative and right-wing forces gained ground (Meriläinen & Mitrunen, 2023).

Three mechanisms connect wartime violence to later voting. First, the post-war land reform and the extension of the municipal franchise reduced economic and political inequality, and voters in the hardest-hit areas rewarded the governing center-right parties for these concessions. Second, in “White Finland”, post-war narratives written from the victors’ perspective portrayed the Reds as dangerous and illegitimate, fueling a backlash against parties associated with them. Third, executions and deaths in prison camps mechanically shrank the pool of left-wing supporters. A violent class conflict thus left a long-run imprint not as renewed fighting, but as persistent electoral cleavages.

Italy: fascist reactivation and resistance legacies

In Italy, twentieth-century legacies pull in two contrasting directions. One runs through the early years of the Fascist movement, built from below through party clubs, street squads and ritualized shows of force. These networks did not vanish in 1945. Where fascist mobilization had sunk deep roots, activists kept a ‘tripod’ of assets alive – knowing what they believed, how to use violence and whom to call on – even when it was costly to show it in public (Costalli et al., 2024).

When Italian democracy later entered the “Years of Lead” in the 1970s and 1980s, these former strongholds were more likely to experience neo-fascist attacks. Violence tended to spike in moments when the state appeared uncertain about how to contain extremism. Fascism’s local infrastructure behaved like a karst river: apparently buried after the war but still flowing underground and ready to surface when political rock layers cracked – old mobilization reappearing decades later in new cycles of right-wing terrorism.

A second Italian legacy runs in the opposite direction. In the final years of the Second World War, armed partisans fought Nazi occupiers and their fascist allies. The units were demobilized after 1945, but in some communities the memory of resistance became a living political resource. Families, partisan associations, cultural groups and municipal officials acted as “memory entrepreneurs”, sustaining local commemorations, naming streets and schools after partisans, and anchoring national narratives in specific sites of arrest, execution or shelter (Cremaschi & Masullo, 2024). More than seventy years later, these same places showed higher support for a grassroots campaign to strengthen Italy’s anti-fascist law. The memory work carried out over decades helped transform the legacy of armed resistance into peaceful, bottom-up civic mobilization in defense of democratic norms.

Portugal: co-optation and blunted protest legacies

Portugal shows how authoritarian co-optation can suppress legacies of opposition. Under the authoritarian Estado Novo regime, repression was paired with corporatist Casas do Povo, run largely by local landowners who distributed benefits, mediated labour relations, offered tightly controlled channels for participation and helped monitor rural workers (Albertus & Schouela, 2024). In the large estates of the south, landless peasants repeatedly mobilised against landlords and the dictatorship, while elsewhere mobilisation was weaker and corporatist structures more entrenched.

After democratization in 1974, areas with a stronger history of protest showed higher support for the Portuguese Communist Party (PCP). But this pattern only appeared where corporatist control had been weak. In municipalities with powerful Casas do Povo, the electoral legacy of past mobilization was far more muted, and those same institutions later helped landowners organize against radical land reform.

Germany: bystanders, perpetrators and accountability

The two German contributions in the Special Issue focus on the legacies of Nazi atrocities but look at different actors. One examines localities exposed to death marches and other visible episodes of mass violence as the regime collapsed. In these places, civilians saw starving prisoners driven through their streets or dying in improvised camps. Decades later, the same areas show lower support for far-right parties and authoritarianism (De Juan et al., 2023). Being forced to confront extreme cruelty against defenceless victims appears to have generated lasting moral and psychological dissonance that undermined residual attachment to the regime’s ideology, pushing politics away from, rather than back toward, authoritarianism.

The second study turns to political elites in post-war West Germany and a key “question of conscience” in 1965–69, when the Bundestag held a free roll-call vote on whether to extend the statute of limitations for Nazi crimes. Members of parliament elected in constituencies with synagogues attacked during the November 1938 pogroms were more likely to support extension; those with Nazi Party pasts were less likely (Charnysh & Riaz, 2023). The geography of violence thus mapped onto an internal divide within the elite between exposed bystanders and former perpetrators, with the bystander experience reshaped in a democratic setting into an active pro-accountability stance.

Israel: victimhood, memory, and ongoing conflict

Israel, a non-European country, offers a different kind of example. In this case, Holocaust legacies are carried less through family histories than through state narratives in the shadow of ongoing conflict. The key divide is between Jewish Israelis more or less exposed to commemorative schooling and public rituals that stress collective persecution and existential threat (Damann, Fachter & Wayne, 2023).

In this context, historical trauma is channelled into ingroup defence rather than empathy: suffering does not translate into solidarity with other victims but underpins hostile attitudes toward perceived enemies, including Palestinians. A legacy of genocide that occurred in European soil thus reaches the present in Israel mainly through education and commemoration, as a defensive collective identity.

Why These Legacies Matter Now

Across these cases, similar episodes of civil war, partisan resistance, authoritarian co-optation, mass atrocities and genocide do not yield a single “legacy”. In some places they produce persistence; in others, regime aversion; in others still, renewed mobilisation or altered outcomes ranging from new party systems to civic campaigns and hardened intergroup boundaries (Ruggeri, Balcells & Justino, 2025). The differences turn on who experienced what, how those experiences were organised and narrated, and which mechanisms of fear, guilt, material incentives, organisational networks or memory work carried repression and resistance into later generations.

For debates on democratic backsliding, transitional justice or the politics of memory, the implication is clear: it is not enough to say that “history matters” and stop there. These studies show what it means to be precise about which past matters, for whom and through what, and to ask whether the relationship is one of congruence or alteration, carried mainly by biographies, local organisations or national narratives. Legacies are neither destiny nor decoration, but patterns created and reworked by institutions, organisations and stories. How societies choose to govern, remember and narrate their violent pasts will keep structuring politics long after the last veterans, bystanders and survivors are gone.

The special issue introduction outlines three key forward-looking priorities for legacy research. First, they call for deeper theorization of mechanisms, urging scholars to examine how legacies are transmitted across micro, meso, and macro levels, why some fade while others reactivate, and under what conditions past events continue to shape present politics. Second, they emphasize the need to make legacy trajectories explicit by identifying when and where different legacy types—such as persistency or transformation—should be expected, rather than assuming automatic continuity. Third, they highlight the importance of attending to multiple levels of analysis, noting that mechanisms operate through individuals and families, communities and organizations, and broader institutional or global forces. Understanding where legacies “live” is crucial for explaining their variation across time and space.

Laia Balcells is the Gallagher Family Professor of Government at Georgetown University. Her research explores the causes and consequences of political violence and repression, warfare dynamics, nationalism and ethnic conflict, and transitional justice. She is the author of Rivalry and Revenge: the Politics of Violence During Civil War (Cambridge University Press, 2017).

Professor Patricia Justino is Director Designate of the United Nations University World Institute for Development Economics Research (UNU-WIDER). Her research focuses on inequality, governance, and inclusive development, particularly the links between political violence, social policy, and long-term development outcomes. She is co-founder of the Households in Conflict Network (HiCN) and a Professorial Fellow (on leave) at the Institute of Development Studies (IDS). She holds an MPhil in Economics from the University of Cambridge and a PhD from the University of London. Her website is https://patriciajustino.net

Andrea Ruggeri is Professor of Political Science at the University of Milan, following a decade at the University of Oxford, where he served as Professor of International Relations and Director of the Centre for International Studies. He also co-coordinates the Italian Standing Group of International Relations, is a Senior Research Associate at Oxford and serves as a Senior Associate Research Fellow at ISPI.

Historians and political scientists have suspected that iron technology may have played a crucial role in this political transformation. Josiah Ober (2015) observed that the availability of iron relative to bronze made it difficult for Iron Age elites to monopolize violence potential. Scheidel (2017) concurs and notes that by the 6th century BCE, a large portion of the Greek-controlled lands had developed a citizen-centered culture characterized by citizen participation in war. Manning (2018) observes that iron led to new warfare and farming techniques, turning the landscape from large cities to dispersed walled villages. Boix (2015) tells the story through looters and producers, and how iron allows the producers a new level of defense. Until now, scholars have not measured the effect or prove the connection using statistical methods.

My study, using digitally constructed maps of the Eastern Mediterranean from the 14th to 8th centuries BCE, provides the first empirical evidence of iron’s revolutionary impact on political power. The findings support the previous research by showing that after regions switched to primarily iron, they experienced over a 100% increase in political fragmentation, more than doubling the number of polities within a given area.

The Bronze Age

In order to appreciate the impact of iron metallurgy, we must examine the world of the late Bronze Age and the political equilibrium iron disturbed. Bronze requires two key components: copper and tin. Copper, while not overly abundant, could be found in several scattered deposits across the Mediterranean. Tin, which was approximately 12% of bronze, was extremely rare with the main deposit in the East being located in modern day Afghanistan. This scarcity led to high prices, creating a perfect monopoly situation for Bronze Age elites.

Bronze weapons and tools were expensive to produce and were restricted to mostly the wealth ruling class and the warriors they chose to outfit. This situation gave Bronze Age elites military superiority over their subject and neighboring smaller polities who may not have been able to meet the weaponry cost. The result was a trend toward large centralized empires that could project military power across vast territories. My study shows that the introduction of bronze is associated with around a 50% decrease in the number of polities within an area, as larger and wealthier polities conquered others.

Blood and Iron

Beginning in 1200 BCE, a new technology started to disseminate from what is today the country of Georgia: iron. Although societies had been aware of and worked with meteoric iron for centuries, learning to work local deposits was revolutionary. Unlike the scarce tin required for bronze, iron ore is to this day one of the most abundant metals on Earth. For the first time in centuries, “ordinary farmers and herdsmen thereby achieved a new formidability in battle” as William McNeill (1982) puts it. Small, poor communities could afford to arm themselves with weaponry comparable to those of the elite.

In Greece, a few Mycenaean kingdoms gave way to hundreds of city-states. Even using conservative estimates there was a 50-fold increase in polities within the mainland and Aegean islands. In the Northern Levant, the Hittite Empire’s collapse was followed a fragmentation region of Neo-Hittite and Aramean city-states who were unable to achieve the territorial scale of their Bronze Age predecessor. In the Southern Levant, the withdrawal of the Egypt left a power vacuum filled by competing kingdoms of Israel and Judah, as well as numerous Phoenician city-states. On the Iranian Plateau, Assyrian records describe the Iranian Medes societies as numerous and small fragmented polities. A stark contrast the unified Bronze Age Assyria.

This was not just about individual weapons being cheaper. Iron technology enabled smaller polities to field armies that could compete with their larger neighbors. The late Bronze Age equilibrium was broken by iron. The number of polities within a region more than doubled in some areas.

Measuring Fragmentation

Although there is previous research and historical qualitative evidence to demonstrate that this link between iron and fragmentation is real, my study focuses on not only on testing the theory, but on measuring this impact. Did it actually happen and if it did, was it actually a large effect?

In order to quantify iron’s impact on fragmentation, I created digital maps showing political boundaries across the Eastern Mediterranean from 1400 to 800 BCE. I then divided the region into grid-cells.

The map data comes from multiple sources. The main source for the MENA region is Brill’s Historical Atlas of the Ancient World. Mycenaean and Iron Age Greece required more specialized studies. Several sources were used to carefully reconstruct Mycenean Greece and later Greek poleis that were politically independent. The data for iron adoption was taken from Peter Turchin and co-authors (2021) and measured when iron became the metal used in weaponry or tools. The benchmark was iron replacing 70-80% of bronze weapons and tools within a region.

Using a two-way fixed effects model, and controlling for geography climate, and competing military technologies, I estimated the effect of iron on fragmentation. Fragmentation was measured as the number of polities per grid-cell in the main specification. The best way to see these results is in the following figure which nicely illustrates the positive impact of iron metallurgy on the number of polities.

Pathways to Fragmentation

But how exactly did iron shatter the equilibrium of political centralization? The cause is multi-faceted, though I present four reinforcing mechanisms:

Increased Interpersonal Violence

Recent research by Baten et al (2023) found that archaeological sites with iron weapons showed higher rates of violent injuries in skeletal remains. Although iron increased political competition at the higher level of polity, it also democratized violence at the individual level, leading to higher rates of violence. This could potentially make polities more difficult to govern as both the elite class and their subjects can kill each other more easily.

Increased Polity Warfare

Although interpersonal violence increased, so too did violence between polities. An analysis of battle records shows an uptick in conflicts as a polity adopts iron. Egypt, which did not reach the iron adoption threshold until 600 BCE, shows relatively low levels of conflict throughout my study. Assyria and the Canaanite region in contract show dramatic increases in warfare after reaching the iron threshold in the 9th century BCE. These upticks are concurrent with empire-building attempts, showing that while iron democratized violence potential, it does not mean new polities were not imperially-minded like their predecessors.

Political Instability

Iron democratizing warfare leads to political instability. Tracking ruler turnover in Babylon, Assyria, and mythological records of Athens shows a striking pattern. All three polities saw an increase in leadership turnover per century following iron adoption. When violence becomes cheaper, changing an unpopular ruler becomes more feasible. Increased political instability also works against imperial expansion, leading to a more fragmented landscape of smaller polities.

Population Growth

Iron adoption not only lowered the cost of weapons, but also that of agricultural tools that enabled agricultural improvements in settlements of previously marginal areas. Comparing population numbers before and after iron shows a remarkable increase in population. Increased population means that communities could field larger armies. A larger army could support independent polities organization in areas previously ruled by foreign entities.

Why should we care?

The ancient iron revolution is not just ancient history. It fundamentally reshaped the political landscape throughout the ancient world. The fragmented city-states of Greece laid the groundwork for innovations in governance. The competing kingdoms of Israel and Judah created one of the most enduring religions the world has seen. Iron changed development in the ancient world.

Understanding how military technologies affect political power remains as relevant today as it did in 1200 BCE. Just as iron democratized violence in the ancient world, modern technologies reshape the balance of power between states and between individuals within states. Technologies being developed now can reshape the political landscape of the world.

Identity politics plays a central role in contemporary political divisions—shaping voting behavior, social preferences, and conflict. Notably, these patterns often contradict material self-interest. For instance, poor white voters in the U.S. often oppose redistribution—despite being its likely beneficiaries (Alesina and Glaeser, 2004). This raises two fundamental questions: What role do economic incentives play in shaping identity politics and racial hate? And, if not economics, what forces drive these behaviors?

Slaveocracy: Elite Capture and the Support for Slavery

To address these questions, we study slavery—arguably the most foundational exclusionary institution in U.S. history and one that continues to shape political identity today (Acharya et al., 2016). While economic historians have thoroughly documented the material incentives that made slavery profitable for slaveowners (Fogel, 1989; Wright, 2006; Olmstead and Rhode, 2010), a major puzzle remains: why did the broader white population—most of whom did not own slaves—so staunchly support the institution? Before the Civil War (1861–1865), 75% of Southern white men were non-slaveholders, yet they defended a system that undermined regional development, depressed investment in human capital, slowed technological progress, and concentrated wealth in the hands of a small elite (Sokoloff and Engerman, 2000; Nunn, 2008; Wright, 2022; Hornbeck and Logan, 2023). More generally, why do extractive institutions that benefit only a narrow elite persist, even under democratic rule?

In new research with Federico Masera, we study how elite economic incentives shaped broader political behavior. We focus on slavery’s profitability—driven largely by cotton—and model how changes in agricultural comparative advantage during Westward Expansion altered elite interests. As the population moved westward, the colonization of new land shifted counties’ comparative advantages in the use of slave labor. Figure 1 shows the geographic variation of changes in comparative advantage between 1810 and 1860 along with the Westward Expansion. Exploiting this variation, we estimate how changes in the returns from slavery translated into political support in Congress. We find a large effect of slave profitability on support for the institution. Interestingly, the effects are too large to be explained by changes in elite voting behavior alone. On the contrary, only about 30% of the change in pro-slavery votes can be attributed to slaveholders. This suggests that economic motives extended beyond the planter class and motivates our analysis of the mechanisms that allowed the planter elite to secure the support of the broader white population.

Elite Capture at Work

We find evidence that planter presence was critical in shaping political behavior. Using a sample of linked individuals between 1850 and 1860, we show that only the planter elite migrated out of counties where the return from slavery was declining—relocating to areas with higher returns (Figure 2). In addition, wages for white laborers were higher in plantation counties, and newspaper content was biased in favor of slavery. As the returns from slavery fell, both wages for white laborers and newspaper content declined (Figure 3).

These results indicate that planters created clientelistic relationships—offering wage premia and security in exchange for loyalty, and threatening retaliation when support was withdrawn. Their control of local media further allowed them to shape political preferences. However, when planters migrated westward—following changes in comparative advantage—support for slavery among non-slaveholders declined. Our analysis indicates that the planter elite captured the political system through their weight in the local economy. As the profitability of slavery declined, planters relocated to more productive land, severing patron-client relationships with local white laborers, decreasing their investments in political persuasion through pro-slavery content in newspapers, and creating new opportunities for non-slaveholders to acquire land and attain economic independence. Because access to wealth required land acquisition—a path blocked by the planters’ monopolization of the most valuable agricultural land (Wright, 1978)—this pattern diminished the income available to white laborers tied to the slave economy while simultaneously opening new opportunities outside of it, ultimately eroding support for slavery. The results align with historical evidence indicating that political consensus in the slave economies was the result of a mix of coercion and incentives (Genovese, 1975; Genovese, 1988; Bolton, 1994; Merritt, 2017). More broadly, the paper shows how even under democratic regimes, the concentration of economic power allows elites to influence the political process and sustain institutions that are morally repugnant, economically inefficient, and serving only their own interests.

The Civil War and Racial Hate in the U.S. South

Results in Masera and Rosenberg (2024) indicate that support for slavery was rooted in economic incentives. When material conditions were removed, political behavior shifted rapidly. This conclusion, however, raises the question of why racial animosity persisted long after emancipation. From an economic perspective, because discrimination is inefficient, it should vanish in competitive markets (Becker, 1957). Yet, while emancipation removed the material foundation of slavery, segregation, violence, and hate endured.

To explain the geography and persistence of racial hate in the postbellum period, Walker, Rosenberg, and Masera (2025) investigate the role of the Civil War in the formation and transmission of a social identity that sustained segregation and racial hate in the Southern United States throughout the twentieth century. We show that the war experience played a critical role in shaping collective identity in the South. While the war led to the emancipation of the Black American population, it also generated a narrative of loss and resentment that became central to white Southern identity, tied to opposition to emancipation and the memory of the Confederacy. We construct a measure of local battle exposure and show that communities more affected by the war exhibited stronger post-war support for segregationist candidates, committed more lynchings, and engaged in more white supremacist activity. Figure 4 shows the main result.

The Formation of Social Identity

To understand the war’s effect, we compare towns with similar overall battle exposure but whose soldiers fought in battles with different characteristics. Exposure to the largest and most renowned battles produced effects nearly three times greater than less prominent but equally deadly engagements. Battles involving Black troops had effects eight times larger than those fought only by white soldiers. These findings highlight two channels: the symbolic power of shared experiences and the identification of Black soldiers as the enemy. We then turn to remembrance, a crucial step in forging collective identity (Anderson, 2006; Zubrzycki and Wózny, 2020). Cemeteries where communities honored fallen soldiers, organizations such as the United Daughters of the Confederacy (Cox, 2019), and white supremacist groups like the Ku Klux Klan (Davis, 2023) played key roles in fostering an environment that sustained racially discriminatory social norms.

We then ask whether collective memory was shaped primarily by elites or grassroots actors (Halbwachs, 2020; Wagoner, 2020). Examining the initiators of Confederate memorials, we find that the families of Confederate soldiers—rather than wealthy individuals or local institutions—were primarily responsible for erecting monuments in communities with higher battle exposure. These results suggest that collective memory was institutionalized from the bottom up. Confederate families led efforts to memorialize the war, building monuments, shaping school curricula, and transmitting identity within households. Crucially, this process was not driven by planter elites. Our evidence shows that the effect of battle exposure does not depend on the size of the formerly enslaved population or the presence of large planters—factors that were key determinants of pro-slavery sentiment in the antebellum South (Masera and Rosenberg, 2024).

Mechanisms of Transmission and Today’s Legacy

Last, we contrast two mechanisms of transmission: one in which Confederate culture persisted through geographically rooted elements, such as landmarks and local institutions reinforcing collective memory, and another in which it was sustained through the intergenerational transmission of individual preferences within families. Comparing a measure of battle exposure that captures family-level exposure to the war to our baseline measure, we find that both channels played an important role, with statistically equivalent effects. Today, the war’s effects persist in hate crimes, police violence, and white supremacist rallies, indicating that the war’s memory still shapes racial animus and collective identity.

Conclusions

These two articles together demonstrate the complex relationship between economic structure and political preferences. Masera and Rosenberg (2024) show that the profitability of slavery, embedded in a democratic system, created strong incentives for elites to build broad support for the institution through clientelistic relationships based on their economic weight and control of local media. When planters migrated, these ties were severed, opening opportunities outside the slave economy for poor whites, who subsequently reduced their support for slavery. This suggests that, absent material incentives, racial animosity might have withered away. Instead, the Civil War transformed racial hostility into a durable identity. This new identity emerged from collective mourning, institutionalized by families of fallen soldiers through associations and memorials, transmitted within households, and grounded locally through organizations and physical markers. These monuments celebrated a mythical past, redeeming the dead by inscribing their sacrifice in a symbolic network (Zizek, 2009). Surprisingly, this process of memorialization was largely bottom-up and yet, somewhat paradoxically, elites continued to benefit—leveraging violence and hostility to maintain control over the emancipated Black population. While racial identities were rooted in the economic structures of slavery, the war reshaped their geography, crystallizing animosity into a lasting identity that extended beyond traditional slaveholding regions.

Michele Rosenberg is an Assistant Professor of Economics at the University of Essex. His research spans economic history, political economy, and development. His work investigates the forces that promote or hinder economic progress and political emancipation, focusing on conflict, collective action, and social norms. Michele was previously a postdoctoral research fellow at Northwestern University and at the Institute for Advanced Study in Toulouse (IAST).

Jerik Cruz, PhD candidate, MIT (https://jerikdcruz.github.io/)

JMP: World Wide Webs: How Migrant Networks and Porous Bureaucracies Forged the Knowledge Economy in the Global South (link to paper)

A long political economy tradition argues that centralized states deploying concerted industrial policies are crucial for developing productive industries. Yet developing countries that have emerged as major exporters of knowledge-based services (e.g. software/R&D/AI services) have often lacked these state structures. I advance a new theory of how the rise of these services-exporting hubs have been driven by skilled migrant networks engaging with porous and dispersed bureaucracies. These structures foster bureaucrats’ connectedness to peripheral entrepreneurial networks, allowing policymakers to leverage distributed tacit knowledge held by migrant co-nationals in processes of fine-grained collaboration. I test this argument using first-ever, agency-level datasets of industrial policy bureaucracies covering all GATT/WTO members since 1989, and historical process-tracing of the Philippines’ and Malaysia’s diverging knowledge economy transitions based on 50 elite interviews. My results challenge a vast literature underscoring “Weberian” bureaucracies and autonomous “developmental states” as prerequisites for structural transformation in the era of knowledge-based capitalism.

Dmitri Kofanov, Postdoc at the Center for Governance and Markets, University of Pittsburgh (https://sites.google.com/view/dmitrii-kofanov)

JMP: Local Democratization and Public Finance - R&R, American Political Science Review (link to paper)

This paper examines the influence of subnational representative institutions on fiscal outcomes and public goods provision, leveraging an understudied large-scale institutional innovation: the introduction of elected self-government in the cities of the Russian Empire from the 1870s. Applying a staggered difference-in-differences design to novel city-level panel datasets, I find that the reform had a large positive impact: it increased revenues and expenditures by 20–50%, enabling an expanded allocation of funds to education, healthcare, fire protection, and other areas, and significantly increased the number of primary schools. The effect size depended on the presence of commercial classes in the city population. Disaggregated budgets suggest that the outcomes were shaped by the preferences of both state authorities and local communities. These results demonstrate that in an authoritarian setting, a reform bringing even limited representation and autonomy at the local level can empower new social groups and promote development.

Daniel Lowery, PhD Candidate, Harvard University (https://www.daniellowery.net/)

JMP: Unmaking the State: Succession Conflict, State Building, and Long-Run Political Development (link to paper)

Internal conflicts impede long-run state development. This paper argues that succession conflicts—violent struggles over the throne—fractured elite coalitions, disrupted bureaucratic expansion, and undermined the state’s ability to monopolize violence or wage external war. These conflicts over the transfer of power had persistent effects. States with more frequent historical succession conflicts exhibit persistently lower fiscal capacity, less effective governments, and more neopatrimonial governance. Using original data on over 2,300 reigns across 115 monarchies from 1000 to 1800, I construct a measure of historical succession conflict exposure and link it to modern outcomes via spatial crosswalks. Instrumental variable estimates in early modern Europe suggest a plausibly causal relationship. The results reveal how the universal problem of succession generated institutional breakdown and help explain durable cross-national variation in state capacity today.

Natalia Vasilenok, PhD Candidate, Stanford University (https://nvasilenok.github.io/)

JMP 1: Reading Orwell in Moscow (link to paper)

In this paper, I measure the effect of conflict on the demand for frames of reference, or heuristics that help individuals explain their social and political environment by means of analogy. To do so, I examine how Russia's full-scale invasion of Ukraine in February 2022 reshaped readership of history and social science books in Russia. Combining roughly 4,000 book abstracts retrieved from the online catalogue of Russia’s largest bookstore chain with data on monthly reading patterns of more than 100,000 users of the most popular Russian-language social reading platform, I find that the invasion prompted an abrupt and substantial increase in readership of books that engage with the experience of life under dictatorship and acquiescence to dictatorial crimes, with a predominant focus on Nazi Germany. I interpret my results as evidence that history books, by offering regime-critical frames of reference, may serve as an outlet for expressing dissent in a repressive authoritarian regime.

JMP 2: Peasant Commune and the Demand for Land Titling in Imperial Russia (link to paper)

Despite their potential economic benefits, land titling reforms around the world often encounter only moderate participation rates. Why do farmers hesitate to claim private land titles? To address this question, this paper examines the historical case of the 1906 land reform in the Russian Empire. For the first time in the country's history, the reform enabled peasants to obtain private titles to plots that had previously been held under communal tenure. Drawing on newly digitized commune-level data from the province of Simbirsk, the paper argues that the perceived benefits and costs of transitioning to private property were shaped by differences in the practice of land reallocation among the members of a commune, known as repartitions. I find that the reform was much less successful in communes where repartitioning had developed as a substitute for the land market. The results suggest that the design of land reforms must account for the incentive structures created by traditional property rights regimes.

The 1933 Soviet famine killed 6 to 10 million people; 60 million starved or witnessed starvation but survived, probably scarred for life. This famine was the second-largest in 20th-century history, after the Great Chinese famine; by the proportion of deaths to population, it was similar to it. 1933 seems far away, but we are talking about our grandparents and great-grandparents living through unimaginable horrors: whole villages depopulated, cannibalism.

Disproportionately many of the victims were Ukrainians: the Soviet republic of Ukraine was just 20% of the population, but accounted for at least 40% of the famine deaths. Ukraine considers the 1933 famine a genocide, while Russia claims that Stalin was “an effective manager” and destroys famine memorials in the occupied territories. The memory of the famine and suffering affects politics and economic development to this day (Rozenas and Zhukov, 2019; Yaremko, 2022; Naumenko, 2024b).

This post summarizes the research Andrei Markevich, Nancy Qian, and I have published studying this famine (Naumenko, 2021; Markevich, Naumenko, and Qian, 2024). The maps below show the distribution of ethnic Ukrainians and 1933 excess mortality; higher mortality occurred where more Ukrainians were present. However, there is still a debate: is this because ethnic Ukrainians were targeted, maybe even by an artificially planned and organized famine, as some claim, or because Ukrainians happened to live in grain-producing areas of the Soviet Union, and grain-producing areas suffered, either from bad weather or bad government policies?

To answer this question, we must take a step back and understand why this famine happened.

Famines in Russian history

First, let’s put the 1933 famine in perspective. How normal, or how abnormal, was it? Russia (the Russian Empire before 1917, the Soviet Union after 1917) was a largely agrarian economy: in 1926, 80% of the population were peasants, and agriculture was roughly 50% of GDP (Cheremukhin et al., 2016). Compared to Europe and the US, agriculture was less productive (Allen, 2003), but after the abolition of serfdom in 1861, yields were increasing steadily until 1913 (Markevich and Zhuravskaya, 2018). World War I, the 1917 Revolution, and the subsequent Civil War interrupted this process; however, by 1928, the rural economy had roughly recovered to the level of 1913 (Markevich and Harrison, 2011). Before 1933, in recent Russian history, there were two large rural famines: in 1892 and in 1922.

The 1892 famine was caused by a severe 1891 drought and was exacerbated by a cholera epidemic; its epicenter was in the fertile but drought-prone Volga region in the south of Russia. This famine killed roughly 0.5 million people (Wheatcroft, 1992).

Civil War destruction and arbitrary grain requisitions during the War Communism era (1918–1921) reduced land under cultivation by more than 30%. A severe 1921 drought with its epicenter in the Volga region further exacerbated the crisis. It is difficult to separate the 1922 famine victims from the WWI and Civil War losses, but estimates range from 3 to 5 million (Adamets, 2003; Wheatcroft, 2017).

Thus, the 1933 famine was unprecedented: without war destruction, with no worse agricultural technology than in 1891, it killed an order of magnitude more people. Moreover, the mortality epicenter was different: in Ukraine, as opposed to the Volga region. Was the weather ten times worse than in 1891? Did something else cause the famine?

Collectivization

In late 1929, Stalin started a comprehensive collectivization campaign. According to official documents, grain-producing southern areas were the first priority for collectivization, while the less-productive north was less of a priority. Under collectivization, peasants were forced to give up their land, livestock, and implements and to work together on collective farms. The harvested grain was put in collective farm granaries, where, after giving the government its share and setting aside seed stocks, fodder, and reserves, the rest was to be divided among the collective farm members. The government tried to prevent collective farm members from keeping private plots and livestock; whatever was left in private property was insufficient for subsistence. Trading of food was banned, and the government used procured grain to feed the urban population and for export.

Collectivization simplified the extraction of grain from the countryside: it is easier to search one collective farm granary instead of a hundred individual households. This essentially meant that the government could put a 100% tax on surplus grain. Historians generally agree that collectivization also destroyed incentives to work: why produce more than subsistence if all surplus is taken away? Consequently, in 1931 and especially in 1932, harvests were lower than the government had planned.

Food Accounting

But was the 1932 harvest decline bad enough to kill millions? The figure below plots grain production from 1913 to 1939 in constant Soviet borders.

Two observations are striking. First, the 1932 harvest is almost twice as large as the 1921 harvest. How is it possible, then, that the 1933 famine had more victims? Second, the 1932 harvest is similar to the 1936 harvest, but there was no mass starvation in 1937. Thus, arguing that there was not enough grain in the country to prevent mass casualties—that, essentially, someone had to die in 1933—is unsound. Moreover, even if we attribute the harvest decline to weather (which I disagree with), weather could not be the main cause of famine; the harvest decline was not bad enough. Somehow, the system of collectivized agriculture-procurement-distribution turned a low harvest into a catastrophe.

Of course, what matters is not only the total amount of grain, but also how it is distributed in the country. Maybe some areas produced too little, and the government tried to help but failed? Markevich, Naumenko, and Qian (2024) show that Ukraine produced less than usual, but this low harvest would have been enough to avoid mass casualties had the government not taken away more than 40%. Consistent with the harvest totals presented in Figure 3, they also argue that the grain taken from Ukraine was not necessary to prevent mass starvation in the rest of the Soviet Union.

Causes of the 1933 famine

Why then did millions die in 1933, both in Ukraine and outside? Here’s what I think happened.

First, collectivization led to a harvest drop, especially in traditionally productive areas that were more intensely collectivized, including Ukraine. Then, insisting that the peasants were sabotaging collectivization and hiding grain, Stalin pushed ahead with procurement, turning a bad harvest into a catastrophe. Grain procurement again targeted areas that usually produced more grain, both because it was difficult for the government to assess how little grain was actually there in 1932, and because it was easier to take grain from more collectivized areas.

Consistent with this explanation, Naumenko (2021) shows that, in Ukraine, harvest drop is not predicted by the weather—therefore, it must have been caused by collectivization. Naumenko (2021) also shows that within Ukraine, districts (relatively small administrative units) with a higher collectivization rate (measured in 1930 or in 1932) experienced higher famine mortality. That is, if you compare districts with similar agroclimatic conditions, similar past productivity, and similar 1932 weather, the more collectivized ones on average suffered more in 1933.

Also consistent with this explanation, using a sample of provinces (relatively large administrative units), Markevich, Naumenko, and Qian (2024) show that 1932 grain production is either unrelated to famine mortality or is positively correlated with it. Had there been a simple harvest failure not exacerbated by procurement, we should see a negative correlation: lower harvest would lead to more deaths. A zero or positive correlation occurred because the government overprocured from the areas that experienced harvest decline relative to normal years but still produced more grain than the North.

The famine was created by government policies, but this does not mean that the famine was created intentionally. Rather, the ban on trade prevented places with food to spare from selling it—if you revealed that you had more than the government suspected, more than enough for subsistence, you would be targeted by harsher collectivization and grain collections next. The government, frantically trying to feed the urban population and not able to collect as much grain as it had planned from traditionally productive areas, left them without subsistence.

Already in 1932, it was obvious that the system was not working. After the famine, procurement quotas were linked to farms’ assigned sown area and thus were made more predictable. Collective farm members were allowed and even encouraged to keep private livestock and gardens (although these were heavily taxed). After paying taxes and delivering the assigned quotas, peasants could sell the remaining produce in markets at free prices. Peasants received most of their income from private plots and livestock and worked in collective farms primarily for the right to keep them (Belov, 1956). Collectivization continued and, eventually, all peasants were collectivized, but this private sector afforded subsistence.

Bias against Ukrainians

If the famine was created by policies (collectivization and procurement) that targeted grain-productive areas, did Ukrainians suffer simply because they lived in productive areas? No. Every policy we can measure was implemented with a bias against ethnic Ukrainians, resulting in severely biased famine mortality.

First, using a sample of districts in Ukraine, Naumenko (2021) shows that from the start of the collectivization campaign, ethnic Ukrainian areas were collectivized more intensely. That is, if you compare districts with similar agroclimatic conditions and similar past productivity, the ones with a higher share of ethnic Ukrainians were, on average, more collectivized in 1930.

Second, using a sample of provinces, Markevich, Naumenko, and Qian (2024) show that 1932 procurement was more intense in areas with a higher share of ethnic Ukrainians. That is, if you compare provinces with similar harvests, on average, more grain was taken from the ones with more ethnic Ukrainians. This resulted in a bias in grain retention (production minus procurement): provinces with more ethnic Ukrainians were left with less food after the 1932 harvest.

Consequently, using both a sample of provinces and districts within provinces, Markevich, Naumenko, and Qian (2024) show that famine mortality was biased against ethnic Ukrainians. That is, if you compare provinces with similar harvests, similar weather, and other characteristics, the ones with more Ukrainians on average had higher famine mortality. To illustrate this, Figure 4 plots average mortality from 1923 to 1940 in Ukraine, Russia, and Belarus: Ukraine experienced a much larger spike in 1933. Figure 5 plots the estimated difference in mortality between Ukrainian and non-Ukrainian provinces after each harvest from 1922 to 1939. This difference remained stable until the harvest of 1931. In 1932 (after the 1931 harvest), Ukrainian areas experienced higher mortality; in 1933 (after the 1932 harvest), the difference was huge. On average, if you move from a province with zero Ukrainians to a province with 100% Ukrainians, famine mortality increased by 50 per 1,000; this is huge given that normal non-famine mortality was around 20 per 1,000.

Markevich, Naumenko, and Qian (2024) show that, similar to provinces, if you compare districts within the same province with similar soil quality and 1932 weather, the ones with more ethnic Ukrainians had higher famine mortality. Moreover, this is true even if we drop Ukraine from the sample and only consider Russia and Belarus—there appears to have been a pervasive bias against ethnic Ukrainians, not just against the Ukrainian Republic.

Finally, Markevich, Naumenko, and Qian (2024) show that, conditional on production, the First Five-Year Plan was planning to procure slightly more from ethnic Ukrainian territories. This does not mean that Stalin planned to create the famine or to starve Ukrainians; rather, it made Ukrainians more vulnerable if the harvest was below the planned level.

Statistical methods do not allow us to attribute the bias. Whether it was driven by local officials or by Stalin and the Politburo, or some combination of the two, we cannot tell. We do know that ethnicity was salient at the time, that in 1918 Ukraine briefly tried to declare independence from Russia, that Stalin worried about “losing Ukraine,” and that Ukrainians were perceived as more resistant to collectivization. We do know that procurement targets and how strongly they were enforced were decided in Moscow. Yet there is no “smoking gun” document where Stalin spells out an intention to target a specific ethnic group. With or without this intention, the result was a devastating famine that killed millions but hit Ukrainians particularly strongly. Nothing can bring back the dead, but we hope this work moves us closer to historical justice.

Author

Natalya Naumenko is an assistant professor in the Depart­ment of Economics at George Mason University. Her research concentrates on economic history and political economy, with a particular focus on Russia and the former Soviet Union. She gradu­ated from Northwestern University in 2018. Before joining Mason, Naumenko was a postdoc at the Political Theory Project at Brown University. Her website is here.

In August 1990, the Prime Minister of India, Vishwanath Pratap Singh, made a historic speech to implement affirmative action to lower castes in central government jobs and higher education — domains that had long been dominated by upper-castes.  The speech, that came be known as “Mandal”, was the start of a profound realignment in Indian politics.¹ The figure below drives home this point. It shows a dramatic rise in the vote-share for the right-wing Bharatiya Janata Party (BJP) before and after the speech in state elections around the country. What caused this shift?

In a recent paper, I offer a different explanation that focuses on the historic legacy of caste inequality in Indian society. The BJP gained more vote-share in places where the preservation of caste status emerged as more salient to upper castes. In these places, the status inequality between upper-caste Brahmans and other castes, going back to the colonial era, had remained a dormant factor in politics until it was activated by a status threat with the Mandal  announcement. In these places, upper caste voters shifted their allegiance to the BJP, and were possibly more persuaded by “ethnic head counting” strategies, because they were more keen to preserve caste status rather than their shared class interests with lower castes.

In my previous blog post I discussed how historic institutions shape contemporary outcomes, long after these institutions have disappeared. More recently, scholars have shown that a similar logic also applies to racial or ethnic inequality. Take Acharya, Blackwell and Sen’s recent research that shows that higher levels of slavery at the county-level before 1865 continues to shape racial attitudes and right-wing support amongst contemporary Southern Whites. They argue that the end of the civil war and the emancipation of Black Americans increased the salience of a white racial identity, and shaped anti-black racism that has persisted through an inter-generational transmission of attitudes on race.

Importantly, they raise three challenges in their study. First, how do we show that ethnic attitudes are not being shaped by the present-day characteristics of ethnic groups such as their size, or their spatial sorting, but instead by the legacies of the past? Second, what are the mechanisms of transmission that connect the past to the present? And third, how do we solve practical challenges in the study of historical legacies? What data will we need to bolster our claims that history matters and how do we establish spatial continuity?

These challenges are even more stark for studying the historic legacies of ethnic cleavages in India. Unlike the well-preserved census records of the 19^thcentury US (the 1891 census being an exception as records burned in a fire), the last comprehensive census survey of caste was done in 1931, and recent attempts to ask Indian citizens about caste have been met with political resistance. Additionally, while the Indian census collects a variety of data at the village, sub-district and district levels in India, elections in India take place in totally different state and national electoral districts created under the auspices of an independent election commission. The task of merging historic census data into present-day census data, and then synching them with electoral districts, therefore, presents a daunting task.

Even if we tackle the data challenges, how do we begin to measure which identity mattered in a multi-ethnic country like India in that pivotal moment?

The historic shadow of status inequality

Caste has been a basis for the organization of Hindu society for around two millennia. The contemporary caste system rests on the ancient four-fold ascriptive classificatory scheme of varnas – Brahman (priests), Kshatriya (warriors), Vaishya (merchants), and Shudra (farmers, laborers and craftsmen). As priests, brahmans had long held a monopoly on educational access in Indian villages. In order to maintain their “ritual purity” brahmans in local, village-level settings often segregated themselves from other groups and maintained exclusive control over educational institutions, water sources, and temples.

The British, in many parts of India, from the very beginning of the colonial era in late 18th century began to rely on the brahman castes as a skilled labor force to staff the government bureaucracy, and in return let them set the agenda on educational spending in the colonial era. Economist Latika Chaudhary shows that between 1857-1920, upper castes often redirected government spending away from primary schooling and into secondary schooling to benefit their children and to restrict education to lower castes. Educational inequality was an especially stark measure of social status inequality in the colonial era.

Fifty years after independence, the specter of educational desegregation following the the affirmative action announcement galvanized upper-caste support for the BJP, a peripheral player in Indian politics up to that point.

To test this, I used newly digitized data at the sub-district (taluk) level from the 1931 colonial census. I use this to create a variety of historic measures including a measure of status inequality: Brahman Dominance— an over representation of brahmans in the literate population compared to their share of the taluk population. I then manually  merged these data into the 1971 census taluks. Then using the interpolation techniques developed by Francesca Jensenius, I was able to merge the administrative and electoral data.

I show that the increase in support for the BJP in state legislative elections was greater in places where upper-caste brahmans had enjoyed greater status dominance in the past. In Figure 2 I show that no such correlation existed between brahman dominance and BJP voting prior to the announcement.

Finally, to test a mechanism – that historic status inequality activated an upper caste identity in preservation of caste status, and that such an identity emerged where historically upper castes had been more dominant — I examine individual-level contemporary survey data. The individual-level regressions show that Brahmans were more likely to vote for the BJP and held more anti-poor and anti-affirmative action views compared to other brahmans in constituencies with higher levels of brahman dominance in 1931.

¹ The “Mandal announcement” referred to the recommendations of the the Mandal Commission, or the Socially and Educationally Backward Classes Commission (SEBC) established in 1979 with a mandate to “identify the socially or educationally backward classes” of India.

Subscribe now

Germany has long been considered the prime example of this model. From Barrington Moore’s landmark study Social Origins of Dictatorship and Democracy in the late 1960s to recent work in political science (Ziblatt 2008, Montalbo 2023, Emmenegger et al. 2024), the story has remained consistent: the strength of aristocratic landowners in rural Prussia underpinned conservative authoritarianism and helped doom the Weimar Republic.

But does this contrast hold up? And does rural inequality always breed authoritarianism? In new research (Bengtsson and Kersting 2025), we revisit these questions with a different perspective. Rather than seeing land inequality solely as a source of elite repression, we consider an alternative tradition, building on scholars such as Dietrich Rueschemeyer, Evelyn Huber Stevens, and John D. Stepens in Capitalist Development and Democracy (1992) and Gregory Luebbert in Liberalism, Fascism, or Social Democracy (1991). In our framework, building on these works and others, stark disparities in rural wealth can trigger class conflict and radicalization—particularly if landowners fail to exercise social control over laborers and peasants. Where elite hegemony is weak, inequality may generate not passivity but politicization, feeding the growth of the organized Left.

To test these competing ideas, we turn to two canonical cases: Germany and Sweden in the late 19th and early 20th centuries. Germany is the country case par excellence of the landlordism-authoritarianism model, while the less studied Sweden typically in comparisons is portrayed as a relatively egalitarian, farmer-dominated society which managed a peaceful transition to democracy and forged enduring class compromises. Thus, these two countries are often seen as opposites, both in terms of social structure and political development.

Land inequality in Prussia and Sweden

To begin with, we revisit the basic empirical assumption that underpins most of the historical narrative: that Sweden had a more equal agrarian structure than Germany. This assumption is widespread across both historical and political science literatures. Yet the data tell a different story. Using detailed historical records, we compare landholding structures in Sweden (1902) and Prussia (1882). One common indicator of landlord dominance is the share of farms larger than 100 hectares (shown in Figure 1). By this measure, Sweden was not less unequal than Prussia; if anything, it was slightly more so. Comparisons on other indicators, such as the Gini coefficient of land ownership, reinforce this view. In the decades around 1900, the Swedish countryside was not an egalitarian exception but part of a broader European pattern of agrarian inequality. This finding is in line with other recent research investigating income inequality. These findings are important in their own right, as they call into question a central tenet of received wisdom. But what matters more is how this inequality translated into political behavior.

Land inequality and political outcomes

If the classic landlord-dominance theory is correct, we should expect that more unequal areas would have higher support for conservative parties, lower support for democratic or leftist parties, and lower voter turnout—reflecting social control by landlords and political exclusion of the working class. But when we look at the data from Germany and Sweden, we find almost the exact opposite.

We examine how land inequality relates to three political outcomes: vote shares for conservative parties, vote shares for Nazi parties, and voter turnout. Using a repeated cross-section model, we regress these outcomes on land inequality, controlling for agricultural employment (to account for rural structure) and other confounders. Election fixed effects account for national trends, and province or county fixed effects address unobserved regional differences. Our main coefficient captures the association between land inequality and political behavior, with robustness checks including controls for soil quality and agricultural productivity.

In Prussia, electoral support for the conservative party was not higher in more unequal rural areas. In Sweden, conservative parties were weaker in districts with greater land inequality. In both countries, land inequality was positively associated with voter turnout—suggesting that political engagement, not apathy or repression, was higher where rural elites were stronger on paper (see Figure 2).

Towards some explanations

This pattern points toward a different political dynamic: one in which inequality fueled popular mobilization rather than elite domination. In Sweden, this dynamic is especially clear. Areas with higher land inequality also saw higher rates of trade union membership, greater support for Socialist parties, and stronger turnout among laborers. Far from being politically dormant, these regions were strongholds of left-wing mobilization. The sharp class divisions created by inequality appear to have politicized rural workers, pushing them into the electoral arena and into labor organizations.

In Prussia, the mechanisms seem somewhat different. Rather than organizing politically, many workers appear to have responded to inequality through migration. Counties with higher levels of land concentration had higher rates of out-migration, especially to urban areas and abroad. In effect, people resisted landlord power not by voting against it, but by leaving. This “exit” pattern weakened the social base of rural elites and reduced their long-term political influence.

Implications

These findings complicate the standard narrative that links rural inequality to authoritarianism. If anything, they suggest that land inequality did not translate to higher levels of anti-democratic attitudes. The persistence of authoritarianism in Germany, or the absence of it in Sweden, cannot be fully explained by differences in landholding patterns. What mattered more was - as suggested previously for example by Daniel Ziblatt’s comparison of Conservative parties in Britain and Germany (Conservative Parties and the Birth of Democracy, 2017) - how political institutions shaped and channeled the effects of inequality—either into democratic mobilization or into breakdown and conflict.

What does all this mean for how we think about landlords and democracy? At a minimum, our findings challenge the idea that landlord dominance inevitably erodes democratic prospects. While classic theories have portrayed powerful rural elites as obstacles to democracy, our analysis suggests the relationship is more complex. Agrarian inequality does not mechanically translate into political repression or authoritarian sentiment. In fact, in both Sweden and Germany, the most unequal areas often displayed signs of political engagement—whether through mobilization, resistance, or migration. These dynamics complicate the simple story that inequality breeds authoritarianism and point instead to a more contingent, context-dependent relationship between rural class structure and political development.

Of course, our study focuses specifically on electoral politics and mass political behavior. We do not examine other important aspects of elite influence, such as control over the bureaucracy, the military, or the judiciary—dimensions that have long been central to accounts of Germany’s so-called Sonderweg. These elements of aristocratic power deserve further scrutiny. But we would argue that electoral politics remains a vital part of the puzzle, especially in light of the growing literature in historical political economy that places voting behavior and referendums at the center of analysis. When it comes to understanding how land inequality shapes democracy from below, it is crucial to study not just what elites do, but how ordinary people respond.

Ultimately, our results call for a more nuanced model of how rural inequality feeds into political life. As scholars like Luebbert have long argued, economic control does not automatically equal political control. Whether landlords succeed in shaping outcomes depends on the strategies they pursue, the openness of institutions, and the capacity of other actors—especially the working class—to organize and push back. In this light, the comparison between Sweden and Germany offers a striking lesson: similar rural structures can produce different political trajectories. What made the difference was not land inequality itself, but how it was mediated by institutions, historical legacies, and bottom-up mobilization.

Authors

Erik Bengtsson is senior lecturer at the University of Lund. His website is https://sites.google.com/view/bengtsson/start

Felix Kersting is a Postdoctoral Researcher at Humboldt-Universität zu Berlin. His website is https://felixkersting.mystrikingly.com/

Between 1792 and 1815 war raged across land and sea. During that quarter-century, Revolutionary and Napoleonic France fought on five continents in what was arguably the first truly global conflict. At the center of the shifting alliances opposing France stood Britain, which fought French expansion with blood and treasure. The sacrifice paid off: when Napoleon surrendered at Waterloo, Britain emerged as the great power of the century.

As important as the war was to cement British global dominance, its most profound effect may have been subtler. Over those twenty-three years, hundreds of thousands of British men left to fight Napoleon, creating acute labor shortages at home. British entrepreneurs responded to these shortages with new, labor-saving machines, accelerating technological progress and growth (cf. Acemoglu 2002). In short, the men of Wellington and Nelson made Britain great—not only in the ways they imagined.

Why it matters – the debate on the Industrial Revolution

The idea that labor scarcity spurred technological progress during the Industrial Revolution is not new. Allen (2009) argued that Britain’s relative factor prices made the Industrial Revolution economically viable: British coal was abundant and cheap, while British labor scarce and dear. It made sense then for British entrepreneurs to introduce machines that used lots of coal but little labor.

While intuitive, the theory has not gone unchallenged. Mokyr (2021) has suggested that high wages reflected the exceptional productivity of British workers, and that—by international standards—British labor was in fact not expensive in real terms. Along with Kelly and Ó Grada (2023), he further argues that what really set Britain apart was the quality of its workforce, as the country was unusually rich in skilled mechanics who developed and maintained new, complex machines.

Adjudicating between these competing views is hard: only one country industrialized first, and comparing Britain to its contemporaries hardly leads to credible quantitative analysis.

Taking the grand theories to the micro-data

In “Fighting for Growth” (2024), Joachim Voth, Alex Trew and I set out to test these grand theories of the Industrial Revolution. Rather than comparing countries, we ask whether local supply of labor and skill can explain the different pace of technological progress within Britain. Our focus is the period of the wars against France, when military recruitment pulled vast numbers of men from the civilian labor force. In 1809, the army and militia numbered 300’000 men, with another 141’000 serving aboard Royal Navy ships (Rodger 2004). Together they accounted for nearly 15 percent of England and Wales male, prime-age population of 1811. To put that in perspective, Napoleon’s Grand Armée, relative to France’s population, was only about one-third as large. Importantly, British recruitment was highly uneven: some regions sent many more men than others (Figure 1). With low labor mobility (Smith 1776), these localized shocks created substantial regional variation in labor costs.

We find clear evidence of a tightening labor market in areas with high recruitment, both in primary sources and in historical data. Writing from Yorkshire, Tuke (1800) observed: “an advance from twenty to twenty-five per cent has generally taken place [in wages], arising […] from the great consumption of men in the navy and army, and consequent present extreme scarcity of hands for agricultural labour.” The situation was similar in Hereford, where “the population is so much thinned by the levies and operations of war, that the farmer in particular, has but little opportunity of selection” (Duncumb 1805). Systematic analysis of wage data—collected from over 20,000 pages of Parliamentary reports—confirms a strong positive association between recruitment and wages in both summer and winter (Figure 2). In some regions, rising labor costs were large enough to push entrepreneurs to experiment with labor-saving technologies.

Technology in the fields

We study how new technology spread through British agriculture. This focus is useful for two reasons. First, agriculture became one of the most dynamic sectors of the economy during these years, adopting machines—like mechanical threshers—that had existed for decades but had failed to gain traction. Second, unlike manufacturing which was geographically concentrated, farmers cultivated almost every corner of England. Thus, a study of agricultural technology can exploit the full geographic variation in recruitment and provide evidence that war-induced labor shortages stimulated technology adoption.

The link between war, labor shortages and technology adoption was clear to contemporaries and visible in the data. Writing from the South of England, Stevenson (1812) noted that “[a] considerable number of thrashing [sic] machines have been erected in this county […] the principal inducement for using them is a scarcity of labourers, which, in a state of warfare, may be expected to be felt most in the maritime districts.” To move beyond anecdotes, we collected detailed data from historical newspapers advertising farm sales and leases. We read more than 20’000 such articles, and whenever one mentioned a labor-saving machine, we geolocated it and added it to our data. The left panel of Figure 3 shows where these machines appeared, while the right one documents a strong, positive association between recruitment and adoption. This evidence provides initial support to the idea that wartime labor scarcity pushed farmers to substitute workers with machines—but is the relationship causal?

Did recruitment cause technology adoption?

Correlation is not causation, and recruitment may rise and fall with adoption without causing it. For one, labor-replacing machines might have displaced rural workers, worsening their employment opportunities and encouraging them to enlist. In other words, causation may run in the opposite direction, from adoption to recruitment. Additionally, a third factor—such as proximity to urban centers—might drive both higher recruitment and faster adoption by easing access to new ideas and potential recruits. In short, granular data alone cannot establish causality.

To make progress, we focus on naval recruitment, which was both massive and constrained by natural factors. Royal Navy wartime demand for men reached beyond experienced seafarers: during the Napoleonic Wars the Navy faced what Dancy (2015) calls a “problem of arithmetic,” as Britain lacked enough seamen to crew its thousand ships. To fill the gap, the Navy recruited hundreds of thousands of inexperienced “landsmen:” according to our back-of-the-napkin math, the number may have reached 280’000. But naval recruitment is compelling not only because of its scale: it was also determined by external constraints that provide a plausible instrument for identifying causal effects.

The reach of the Royal Navy

We identify the causal effect of recruitment using variation in coastal access. In the 18^th century, every Royal Navy captain was responsible for raising his crew from British coastal towns. Whether captains could visit a town depended however on the seabed offshore: the Navy’s largest ships rose high above the waterline to accommodate three decks of cannons, and they needed as much water underneath for the keels that kept them stable. While this may suggest exploiting access to deep sea as an instrument for naval recruitment, proximity to deep sea also correlates with distance to the coast, and being close to the coast may directly affect the local labor market. To isolate quasi-random variation in naval recruitment, we thus focus on a narrow strip of land within 15 km of the British shoreline and use distance to deep sea as our instrument (Figure 4, left). Within this coastal sample, distance to the deep sea is unrelated with many local characteristics, including the presence of commercial (i.e. non-naval) ports. In short, within the coastal sample, distance to the deep sea appears as good as randomly assigned, and it qualifies as a valid instrument for Royal Navy recruitment.

To illustrate the logic of our instrumental variable strategy, consider two ancient hundreds in Norfolk: Clackclose and Smithdon (Figure 4, right). Both lie on fertile coastal land near the mouth of the Great Ouse: just 32 km separate them. In 1801-11, they were home to around 1’000 people each—mostly rural workers—and were similar along all observable characteristics. All except one: Smithdon faces the deep sea of East Anglia while Clackclose lies on shallow waters of the Great Ouse estuary, 16 times as far from the deep sea (2.4 vs 39 km). The Royal Navy recruited in Smithdon but not in Clockclose, resulting in greater gender imbalances (1.05 vs 1 women per men) as well as in faster technology adoption: we find twice as many labor saving machines in Smithdon (14 vs 7).

Our instrumental variable strategy generalizes this logic across the entire coastal sample. We find that areas with deep sea access experienced more naval recruitment and faster technology adoption: a doubling of recruitment led to one extra machine. Since seabed depth is plausibly unrelated to local labor market dynamics in otherwise similar coastal areas, the link between recruitment, labor scarcity, and adoption is likely causal.

The role of skilled labor

Was labor scarcity all that Britain needed to take off? Machines could replace the men fighting Napoleon—but they still had to be built and maintained. Kelly, Ó Grada & Mokyr (2023) argue that Britain had an unusually large supply of expert mechanics with the skills needed to assemble and service these machines. To measure the role of these expert mechanics, we look at the years leading to the Napoleonic Wars and count how many young men apprenticed to become millwrights, wheelwrights and watchmakers. These apprentices learnt mechanical skills essential to build the complex machines we study: their presence, in theory, should facilitate technology adoption. This is what we find: areas with at least one mechanic adopted significantly more than those without (Figure 5, left). More strikingly, these mechanics also magnified the impact of recruitment on adoption (Figure 5, right): in areas with skilled mechanics, labor scarcity led to much more adoption than areas without it. In sum, labor scarcity and skill abundance both promoted technological progress – indeed, they reinforced each other.

The first industrial country

Why did England industrialize? It is a mistake to believe that a single factor can account for Britain’s take-off (Koyama & Rubin 2022). Our work provides support to the two most prominent theories of the Industrial Revolution and shows that labor scarcity and skill abundance complemented each other in promoting technology diffusion. Naturally, technological progress started earlier than 1800 and was fastest in manufacturing sectors such as textiles. However, the forces we identify were at play throughout the 18^th century, when Britain fought a war one year every three. Technological progress during the labor shortages of the Napoleonic Wars may thus explain why the country that fought the most, was also the country that industrialized first.

References

Acemoglu, D. (2002). Directed technical change. The review of economic studies, 69(4), 781-809.

Allen, R. C. (2009). The British industrial revolution in global perspective. Cambridge University Press.

Dancy, J. R. (2015). The myth of the press gang: volunteers, impressment and the naval manpower problem in the late eighteenth century. Boydell & Brewer Ltd.

Duncumb, J. (1805). General View of the Agriculture of the County of Hereford. London Phillips.

EMODnet (2018). EMODnet Digital Bathymetry (DTM 2018).

Kelly, M., Mokyr, J., & Ó Gráda, C. (2023). The mechanics of the Industrial Revolution. Journal of Political Economy, 131(1), 59-94.

Koyama, M., & Rubin, J. (2022). How the world became rich: The historical origins of economic growth. John Wiley & Sons.

Mokyr, J. (2021). The holy land of industrialism: Rethinking the Industrial Revolution. Journal of the British Academy, 9, 223-47.

Rodger, N. A. M. (2004). The command of the ocean: A naval history of Britain, 1649–1815. Penguin Books.

Smith, A. (1776). An inquiry into the nature and causes of the wealth of nations. W. Strahan and T. Cadell.

Stevenson, W. (1812). General View of the Agriculture of the County of Dorset. London McMillan.

Tuke, J. (1800). General View of the Agriculture of the North Riding of Yorkshire. London McMillan.

Voth, H. J., Caprettini, B., & Trew, A. (2024). Fighting for growth: Labor scarcity and technological progress during the British industrial revolution. CEPR Discussion Papers, (17881).

Bruno Caprettini is Assistant Professor at the University of St Gallen (Switzerland). He studies long-term growth and political economy, with a particular focus on the Industrial Revolution. He holds a PhD from Universitat Pompeu Fabra (Barcelona).

The Republic’s democratic institutions included universal male suffrage, direct election of all magistrates on an annual basis, and lawmaking exclusively by citizen assemblies. In addition to these familiar elements, the Republic employed more unusual practices such as electing all magistrates in pairs that had the ability to veto each other, and assembly voting based on groups similar to the U.S. Electoral College.

The Republic has been overshadowed by its successor, the Roman Empire, an autocracy ruled by a succession of colorful emperors, many of them well-known even today.

There are significant barriers to studying the Republic because information on its institutions and political practices is not readily available to social scientists. Here I would like to highlight some of its key institutions that should be of interest to political economy scholars, and suggest that studying the Republic may be fruitful for understanding democratic resilience and performance.

Political Institutions

Institutions are a core concern of political economy research, and it’s a central tenet of field that institutions are important for long-run political and economic performance. The Republic did not have a written constitution – its key political institutions were established and sustained by custom, in a way we don’t fully understand. The Republic’s key institutions were:

Assemblies

The source of all political authority in the Republic was the will of the Roman people, as expressed through its citizen assemblies. The assemblies had two main functions: they elected all magistrates and they passed all laws and made key political decisions such as declarations of war or acceptance of peace treaties. The Republic was both a direct democracy and a representative democracy.

All adult male Roman citizens were eligible to vote in the assemblies, provided they were physically present in Rome on election days. Unusually for an ancient state, Rome gave voting rights to former slaves, of whom there were a great number. Also unusual, the Republic gradually granted voting rights to residents of allied cities within its territory.

There were two main assemblies, the Centuriate Assembly that elected the very top magistrates and had authority over capital trials and war declarations; and the Tribal Assembly that elected all other magistrates and passed the laws. All citizens were members of both assemblies. Voting was initially by voice but shifted to secret ballot in the 2nd century BC.

The assemblies differed in their vote counting rules. Both assemblies used a group voting system – citizens were divided into geographic groups akin to electoral districts – with each district casting one vote based on the majority view of its members. In the Centuriate Assembly, citizens were also divided into property classes in a way that heavily overweighted the wealthy. No wealth classifications were made in the Tribal Assembly.

Unlike contemporary Greek democracies, such as Athens, where amendments from the floor were common and a law could be passed the day it was proposed, the agenda in Roman assemblies was controlled by magistrates. Laws had to be posted publicly weeks before they came to a vote and normally could not be amended without advance notice.

The use of group voting was a distinctive feature of Roman democracy, without obvious parallel in the ancient world. Understanding whether this practice contributed to the stability of Roman democracy is an interesting research question since the contemporaneous Greek cities were notorious for the instability of their assemblies, a problem that did not appear in Rome. One obvious effect of the group voting system was to prevent dominance by urban voters who could easily attend; the interests of rural voters would be reflected in their group vote even if few of them were present.

Magistrates

Roman magistrates (executives) had substantial power and ability to exercise initiative, but Roman institutions took great care to fragment and check their power. All magistrates – more than 60 in total – were elected annually, most for one-year terms, and incumbents were generally ineligible for re-election for a span of years that varied by office. To protect against executive overreach, almost every offices were held by two (or more) men concurrently, who could veto each others’ action. All executive actions could also be vetoed by any one of 10 tribunes. The Republic’s extensive system of checks and balances feels almost modern, and investigating their interplay and role in the state’s longevity could be fruitful.

The apex office was that of consuls, who led the Republic’s armies, oversaw its elections, and handled ad hoc administrative issues. Other key magistrates included praetors, who oversaw judicial matters, aediles, who managed public infrastructure and festivals, and quaestors, responsible for financial administration.

A somewhat unusual position was the tribune of the plebs, 10 of whom were elected each year. The tribunes had no executive responsibilities, but they oversaw the Tribal Assembly and thus served as legislative leaders, with the power to introduce laws. They also had the power to veto actions of all magistrates and the Senate.

There were no restrictions on who could run for office, other than minimum age limits that increased for higher offices. We have only fragmentary data on candidates and no data on election returns, but it seems that elections were highly competitive and candidates engaged in yearlong campaigns.

Senate

The Senate was a central part of Roman government, but the most difficult one to understand from a modern perspective since it has no contemporary parallel. It was composed of former magistrates, who were enrolled for life after holding high office. The Senate did not have the power approve laws, instruct magistrates, declare war, or sign treaties.

Nevertheless, the Senate de facto took the lead in formulating much state policy, including financial matters and foreign policy, met with embassies of other states, recommended laws to the assemblies, instructed magistrates, and sometimes regulated policies in other cities controlled by the Republic – and it’s clear that its recommendations were usually followed.

It is highly misleading to think of the Senate as a version of a modern legislature or parliament. It was more like a collection of former high-officeholders who collectively manage much of the state’s business through the collective prestige of the members.

While the Senate had a preeminent place in governance, this was by custom, and there was no disagreement that the ultimate power lay in the assemblies. The senate’s decisions (decrees) could be and sometimes were overruled by the assemblies, and magistrates did not always follow its advice.

It seems quite likely that the Republic’s long-run performance was closely linked to its Senate. Exactly how it functioned in the constitutional system, and how its authority drew on custom is poorly understood, and would be an interesting subject for research. The Republic’s final decades saw a breakdown of the Senate’s authority as political actors abandoned customary deference.

The Historiographical Debate

It was once thought that the Roman Republic was an oligarchy, in which a small number of hereditary families controlled the government and elections were a mere façade. That view still lingers in popular discourse, even though classical historians have systematically refuted almost all of its assumptions in the last half century. Classicists still hesitate to refer to the Republic as a democracy, preferring instead to note its “democratic elements” but functionally, if it were around today, there is little doubt that it would be binned as a democracy (its main anti-democratic feature to modern eyes would probably be its restriction of voting rights to men.)

One way to put the Republic’s democratic institutions in context is with democracy index scores. The table shows democracy index scores that I calculated for the Republic based on the rubrics of Polity5 and the Economist Intelligence Unit. I also calculated scores for classical Athens.

Polity scores are calculated by giving Democracy and Autocracy scores across five dimensions, and then transforming them into a final score that ranges from 0 to 10. The bottom-line index score is POLITY = Democracy – Autocracy, which can range from -10 to 10. The Roman Republic and Athens both score very high on democracy, low on autocracy. Rome’s POLITY score is 9, squarely in the democratic category. Some comparators for 2018 were United States (8), Japan (10), India (9), Switzerland (10), China (-7), and North Korea (-10).

The EIU scores are based on 60 questions, grouped into five broad categories. Within each category scores are normalized to range from 0 to 10, with higher scores the most democratic. Overall, Rome scores 6.57 and Athens 6.66, putting them in the category of “flawed democracies” with Argentina, Brazil, Hungary, and the Philippines. Both scores are dragged down primarily by the political participation category due to male-only voting and the requirement that citizens appear in person to participate.

This post is based on the more detailed working paper, “The Democratic Institutions of the Roman Republic,” available at johnmatsusaka.com and https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5179765

Author

John G. Matsusaka is the Charles F. Sexton Chair in American Enterprise, Professor of Finance and Business Economics, and Executive Director of Initiative and Referendum Institute at the University of Southern California. His most recent book is Let the People Rule: How Direct Democracy Can Meet the Populist Challenge (2020).

Subscribe now

In a previous post, I discussed ways in which recent historical political economy research has advanced the study of state capacity. In a field long dominated by the European state-building experience, studies on China have challenged our prior assumptions about the origins and evolution of state capacity.

The Chinese case offers three key departures from the European case. First, it is possible to build the very apparatus needed for extraction without actual extraction. This challenges a very core idea in the bellicist literature: not only are bureaucratic institutions just the “happy-byproducts” of war, their mere existence doesn’t guarantee rising extraction levels. Second, rulers in the quest to build a centralized bureaucracy faced challenges not just from elites trying to hold an Imperial court in check, but from local level actors in the form of provincial elite networks, peasants, and bureaucrats that shaped the character and quality of the bureaucracy. Third, the Chinese case offers a startling example of a reversal of bureaucratic capacity.

1. Bureaucratic Capacity Before War-Making

A defining feature of Chinese political history is a continuous state bureaucracy since the third century BCE. The examination system—introduced under the Sui dynasty (581–618 AD), expanded during the Tang (618–906 AD), and consolidated in the Song (960–1279 AD)—became the principal route for recruiting officials. Successful candidates served three- to five-year rotations at the capital, provincial, or county levels and were barred from serving in their home districts to minimize local bias. Successive dynasties refined this inherited system.

Stasavage (2021) argues that writing technology’s evolution in the Shang dynasty was crucial for China’s early administrative structures. Using data on bureaucratic presence from 186 societies and geographic distances from centers of early writing, he shows that societies adopting writing systems were more likely to develop state institutions. This finding suggests that technological innovations can precede and enable bureaucratic capacity.

Despite two millennia of a sophisticated bureaucracy and centralized tax system—centuries before Europe—China’s tax revenues began to lag behind European states by the early modern period (Dincecco and Wang 2018; Ma, 2011; Sng, 2014).

2. Early Modern Decline in Chinese Taxation

Until the early 1600s, Chinese tax collections matched or exceeded English revenues relative to GDP. In 1423, the Ming empire raised about 5.5 percent of GDP—over three times England’s rate (Stasavage, 2020). By the mid-nineteenth century, Qing tax revenues fell to 1–2 percent of GDP, compared with 15–20 percent in Tokugawa and early Meiji Japan and 10–15 percent in eighteenth-century England. Historians offer several explanations:

1. Size and Control: Sng (2014) focuses on late Imperial China (c. 1650–1850) and argues that China's large size impeded the ability of the ruler to control bureaucrats.

2. Incentive Structures: The Chinese empire created a set of conditions that weakened imperial incentives to engages in extraction. The Qing empire focused on territorial expansion by co-opting enemy territories, and reduced the need to prepare for sudden external attacks. Unlike Europe, the Chinese state development had both an absence of external threats and European style institutional constraints which led to weak investments into a tax system where the ruler paid bureaucrats good wages in return for tax revenues. (Ma, 2019).

3. Decentralization: Qing state simply relinquished everyday governance of tax collections, security and even legal matters to local communities and showed little will nor invested in resources that could intervene in sub-county matters (Zhang, 2023).

These studies allude to principal-agent problems where excessive exploitation came not from a autocratic despot, but local officials who enjoyed tremendous discretion. China's low tax capacity could be understood as an equilibrium where given a pre-existing bureaucracy, the ruler was unwilling incite a tax rebellion and also unable to keep the corruption of local officials in check.

3. Strategic Retreat: The 1712 Freeze

Chinese state-builders, however, have always faced the challenge of a large territory; size being a relatively static factor through Chinese history, despite fluctuations across dynasties. A key episode highlights how the drop in taxation was a strategic choice made by Qing rulers. In 1712, after the conclusion of a successful empire-wide cadastral survey, the Kangxi emperor called a halt to any future assessments surveys as he was satisfied with the revenue base that year. As no further surveys were conducted, land tax rates remained frozen at their 1711 level, leading to a rapid decline in fiscal revenue and a stagnation of the administrative bureaucracy. Similar to early 20th century India or 19th century United States, a relatively high capacity state simply gave up its historic bureaucratic advantage in collecting information about economic surplus.

Zhang (2023) argues that the Qing state's deliberate retreat from fiscal capacity was a manifestation of an ``empirical ideology" guiding its approach to taxation. Analyzing elite debates and documents from that era, he notes that elites of the Qing court adopted a pragmatic/utilitarian view that “increasing agricultural taxes will lead to severe economic hardship among the general population, and therefore lead to social unrest and rebellion” (Zhang (2023, pg.17) This belief was partly based on what they thought were the failures of the Ming era---that taxation had disastrous effects on subsistence agriculture. Even when empirical reality began to diverge in the 18th century and agricultural surplus grew, the ideology held sway over fiscal capacity investments and policy, resulting even worsening fiscal outcomes in the 19th century. He (2013) shows that despite experiments in public finance in the 19th cenutry, the Chinese state failed to transform into a modern fiscal state in part due to the continued reliance on decentralized governance arising from a decaying administrative state.

A weakened bureaucratic machinery was further prone to Internal conflicts and reshaped elite–state relations. Bai et al.(2023) study war mobilization efforts to suppress the Taiping Rebellion in 19th century China. The rebellion, that lasted 14 years (1850-1864) was one of the deadliest in human history killing at least 20 million. The paper studies recruitment and mobilization for the Hunan army that under the leadership of a scholar-general Zeng Guofan finally managed to suppress the rebellion. Using records on China’s civil service exam system and marriages, kinship and friendship networks the paper shows that counties with more dense connections to Zeng experienced more casualties in the war. These same counties also saw the rise of new provincial and national-level elites. In a notable twist to Charles Tilly’s famous aphorism – war made the state, and the state made war -- they show that elites made war, and war made elites.

Peng (2022) studies 17th century China and argues that rulers strategically used patronage appointments to secure not just loyalty but to gain trusted bureaucrats in times of conflict. She shows that the Imperial court was more likely to make patronage appointments during times of internal and external crises in order to mitigate risk. Conversely, the court relied on meritocratically chosen bureaucrats during time of peace. In this way, wars possibly weakened the meritocratic state..

Lessons Beyond Europe

1. Institutional Innovation ≠ Extraction: A sophisticated bureaucracy may under-extract when strategic, ideological, or agency considerations prevail.

2. Local and Moral Agency Matter: Provincial elites, low-level bureaucrats, and peasant communities actively shape state-building outcomes, complicating European-focused theories.

3. Capacity Is Reversible: Importantly for all of us observing Trump in action today: deliberate choices—such as the Qing freeze—can undo state strength. Once weakened, building back that capacity is hard and further makes the state prone to capture by warring elites and factions.

This is an excerpt from my Annual Reviews piece, available here

Historical Political Economy (HPE) has gained considerable traction in the social sciences over the past decade. A dedicated journal and its own handbook have greatly contributed to fostering a scholarly community and enhancing the field’s visibility. Yet, these efforts remain heavily centered on scholars and institutions in the United States. Consider the new Oxford Handbook of HPE: Only four of the 65 contributors are based at European institutions. Recognizing this imbalance, the handbook’s editors conclude: “the future frontier of the field likely falls outside of the United States, which is the concern of a disproportionate amount of the current literature”.^[1]

What does it take to advance this frontier and to more firmly establish HPE as a research field in Europe? To develop some answers to this question, we tried to map the current state of HPE in Europe through a survey among researchers. Moreover, we organized a workshop for researchers interested in the establishment of the field together to identify challenges and opportunities.

We fielded our survey in late 2024 and received 59 respondents from HPE researchers working at European Institutions.^[2] In February 2025, we brought together 28 researchers from all over Europe-and beyond-in Hanover, Germany. We were fortunate to receive generous financial support by the Volkswagen Foundation through a new, rather unconventional, funding line designed to support emerging research areas. Instead of the usual paper presentations, we had a full three days to engage in discussions on the philosophical and institutional foundations of the field.

Here, we summarize the main take-aways from our survey and the workshop discussions.

A growing field with weak institutional foundations

A first result of the survey – echoed during the workshop – was wide agreement on the perception that HPE is a growing field in Europe, but one that still lacks institutional foundations. When asked to what extent HPE is currently an established field in Europe on a scale from 0 to 4, the average response was just 1.9.

This lack of institutional foundations is reflected by the fact that very few respondents completely specialize in HPE: Only a quarter of respondents described HPE as their “most important field of research”, whereas more than half classified it as: “One of several important fields.” Similarly, less than 10% of respondents indicated that they teach pure HPE courses. For most respondents, HPE topics are something they have to integrate in courses with a broader thematic focus, and over one-third indicated that they do not teach HPE content at all.

Given this situation, how can HPE be more firmly established in Europe? Our survey results and workshop discussions suggest three key priorities, all related to a core feature of HPE: Interdisciplinarity.

Interdisciplinarity and its limits

Interdisciplinarity is clearly very central to the identity of HPE. Generally, our survey revealed a strong commitment to interdisciplinary engagement, with more than two-thirds of respondents agreeing that “HPE is an interdisciplinary field.” Large majorities also reported to have participated in interdisciplinary conferences and to have worked in interdisciplinary research teams. However, our discussions also showed that interdisciplinarity raises certain challenges for the establishment of a joint institutional setting, especially for building academic networks and for publishing.

Given the limited institutionalization of HPE as a distinct field in Economics or Political Science departments in Europe, junior researchers, in particular, have strong incentives to stay within the confines of their main fields: Economists try to build a network of economists; political scientists and sociologists do the same within their fields. This disciplinary separation makes it difficult for HPE researchers to exchange ideas. The key challenge is to create spaces for interdisciplinary exchange without weakening the links to existing and important communities in everyone’s main fields. Our survey and discussions clearly attest to a strong demand for such spaces. Conferences or smaller workshops that are appealing to individuals from different fields offer a first step for bridging this gap.

The area where the practice of interdisciplinary work was most muted was in publication strategies. More than a third of Economists and half of all Political Scientists reported never having submitted work to journals outside of their discipline. When asked which journals they would recommend sending HPE work to, Political Scientists overwhelmingly recommended journals from their own discipline, most prominently the APSR, CPS, and the JOP. Economists were actually slightly more likely to recommend journals outside of their discipline: While the Economic Journal and the Journal of Economic History were most frequently mentioned, the APSR ranked third. We leave it open whether this means that political science journals are better able to appeal to interdisciplinary audiences or that economists are more interested in publishing outside their field.

During the workshop we discussed the career concerns associated with publishing outside one’s field. A strategy that has been proven successful for some interdisciplinary teams is to agree ex-ante to which displine’s journal joint work will be submitted. Repeated coauthorship and alternating disciplinary journals can unlock the interdisciplinary prisoner’s dilemma, but this is easier for tenured researchers that are able to plan over longer time horizons.

Beyond concrete strategies, many participants emphasized the value of low-stakes exchanges across disciplinary boundaries. While publication practices remain largely siloed, research workshops, joint reading groups, or co-teaching can be a fruitful exercise to start engaging with researchers from other disciplines and identify complementarities. In a field like HPE, where intellectual common ground must often be actively constructed, this kind of regular dialogue may be more important than is typically acknowledged. Sometimes, “just” talking to each other is not a distraction from academic progress – but a prerequisite for it.

An open field, but open to whom?

An open question for HPE is what disciplinary boundaries it can - and should - seek to cross. While it is quite popular to describe HPE as having three constituent parts of History, Politics, and Economics (cf. Figure 1 from Jenkins and Rubin 2024), the actual collaboration with the Historical discipline is very limited. For example, the HPE Handbook mentioned above features only a single historian (Benati/Carugati 2023). By contrast, participation in our survey and workshop suggests that, perhaps, sociology is a discipline that has much more overlap with HPE than has previously been discussed. This certainly concerns the theoretical commitment to hypothesis testing and generalizability. It may also concern methodological commitments, in particular when one takes seriously the call for more qualitative work that some survey participants mentioned. There are also strong overlaps when it comes to research questions in HPE and sociology: When asking for the most important topics within HPE in our survey, among the top answers were many areas which are also central to sociology, such as “nation building”, “identity”, “culture” and “protests”.

Several participants mentioned during the workshop that they were initially unsure what to expect from a three-day event without traditional research presentations and with a strong emphasis on meta-level discussions. What exactly had they signed up for? Would the format even work? As organizers, we arrived in Hannover with similar questions. What if the conversation stalled after just half a day because researchers were more comfortable discussing their own work than engaging in abstract reflections? In the end, these concerns proved entirely unfounded.

Fittingly for a workshop focused on meta-level inquiry, our most important takeaway also lies at that level: to foster meaningful interdisciplinary collaboration among scholars of institutions, it may be especially productive to encourage them to reflect on the institutional settings in which they themselves are embedded.

^[1] See page 3: Jenkins, J. A., & Rubin, J. (2024). Historical Political Economy: What Is it? In: Jenkins, J. A., & Rubin, J. The Oxford Handbook of Historical Political Economy. Oxford University Press.

^[2] Most survey respondents hold PhDs in political science (46%) or economics (42%), primarily from universities in Germany (23%), the United Kingdom (20%), the United States (16%), or Italy (11%). Approximately 60% completed their PhD within the past 10 years and 72% identified as male.

Authors

Bastian Becker is a Postdoctoral Researcher at the Humboldt-University of Berlin. He received his PhD in Political Science from the Central European University. His website is www.beckerbastian.net

Lukas Haffert is associate professor of comparative politics at the University of Geneva. He holds a Ph.D. in Political Science from the Max Planck Institute for the Study of Societies. His website is www.lukashaffert.com

Cathrin Mohr is a Post-Doc at the University of Bonn. She holds a PhD in Economics from LMU Munich. Her website is cathrinmohr.com.

Labor migration to industrial centers was a central force behind changing identities. Classic theorists of modernization and nationalism—Ernest Gellner, Karl Deutsch, and Eric Hobsbawm—highlighted the role of states and schools in shaping mobile workers into national citizens through standardized, homogeneous cultures. But identity transformation also unfolded from below. As people “exchanged the sluggish, dull life of the rural districts for the bustle of the great cities” (Engels, 1845) and were “rescued from the idiocy of rural life,” (Marx and Engels, 1847) they encountered fellow migrants who spoke different dialects and came from varied backgrounds. Communication across these divides required common linguistic and cultural ground. Eugen Weber, in Peasants into Frenchmen, captures this bottom-up process: “Increased mobility and social exchanges played their part in the spread of national language, as they had in the spread of national currency and measures. Industrialization also helped speed the process. In Vosges, for example, the installation of a cotton industry in the 1870s and its expansion after the 1880s all but wiped out the local dialect when country people moved into small industrial centers.”

But how, exactly, did this process unfold? Weber’s account suggests that industrial centers tended to see their local cultures erased. Many scholars have assumed that local identities persisted mainly in the agricultural hinterlands left untouched by industrialization. Yet this claim remains largely untested and theoretically underdeveloped. Industrial growth doesn’t just attract outside labor; it also creates local employment, reducing incentives for native populations to leave. Even when migrants arrive, whether they come from nearby culturally similar areas or more distant, diverse regions depends on the full landscape of migration opportunities. Industrialization in one location may not draw distant migrants if other regions are simultaneously industrializing and competing for labor.

Another unresolved question in the literature is which identities migrants in diverse settings actually adopt. Education, conscription, and other state-led nation-building efforts often impose the language and culture of the political center—a process well-documented across cases, with France serving as the canonical example. Eugen Weber’s account, and more recent empirical work, illustrate how top-down integration can reshape identities (though this is not the only, or even the most effective means of homogenization). But not all European states pursued such centralized assimilationist strategies. In their absence, identity formation begins to resemble a coordination game. As David Laitin has argued in his work on language, identities that are already widely shared become more valuable for communication—and therefore more attractive. Yet coordination games admit multiple equilibria, making it difficult to predict which cultural norms or identities will prevail when heterogeneous groups converge in migration destinations.

In a working paper with Theo Serlin, we develop a theoretical framework to examine how labor migration shapes identity change. Our focus is 19th-century Britain—a case that, while not devoid of nation-building efforts, lacked the centralized identity engineering characteristic of France. Military service was voluntary, and education remained locally administered. We concentrate on the period of the Second Industrial Revolution, when industrial activity not only surged in scale but also changed in geographic scope. New industrial centers emerged beyond the traditional textile-producing regions, including the coalfields of South Wales and Northumberland, as well as the broader region around London, where industry increasingly clustered to capitalize on growing investment from the City. During the same period, people increasingly migrated over longer distances (see Figure 1).

We examine how industrialization reshaped identity by considering individuals’ incentives to choose both where to live and which identities to adopt. We embed these choices in a theoretical model that connects these migration and identity choices to shifts in the spatial distribution of economic activity—and broader processes of modernization, such as declining migration costs. The advantage of this approach to a more standard empirical design that would link local industrialization to local identity outcomes is that it captures general equilibrium effects: what happens in one location depends on conditions and choices across the entire country. This allows us to study the global impact of local changes.

To estimate the parameters of our model, we use rich individual-level data from the 1851 and 1911 censuses, focusing on migration patterns and identity choices—proxied by personal names. We begin by developing a measure for how strongly names reflect regional identities, in two steps. First, we use a data-driven clustering algorithm on surnames in the 1851 census. Without any geographic input, the algorithm recovers spatially distinct groupings across England and Wales—corresponding to recognizable regions like East Anglia, the West Country, and North and South Wales (see Figure 2). These “cultural clusters” are not only spatially coherent; they also predict meaningful behaviors. Migration and intermarriage are more frequent between districts within the same cluster, even when controlling for geographic distance. Next, we construct an empirical measure of how distinctive first names are to each cultural cluster. We validate this measure in multiple ways, showing that names consistently predict behaviors tied to identity, including spoken language, marriage choices, and migration patterns.

Using names as a proxy for identity, we document key patterns of cultural change between 1851 and 1911. We find convergence toward naming practices associated with the Southeast (SE) cultural cluster—the area surrounding London (see Figure 3). Entire cultural clusters disappear from the map. This convergence is especially pronounced in peripheral regions, as measured by distance from London (Figure 4, left). However, industrial development—proxied by the presence of coal—moderates this effect. Peripheral areas gravitated toward the identity of the Southeast only when they lacked industrial growth. Where industrialization did occur, local cultural identities proved more resilient. Industry did lead to cultural change, but only in the center of the country (Figure 4, right).

Our theoretical model explains these observed patterns (see Figure 5). To better understand the mechanisms linking economic development, migration, and cultural change, we conduct counterfactual simulations—effectively asking what cultural outcomes would have looked like had industrial activity remained at its 1851 levels. These exercises show that industrialization fully accounts for the rise of Southeast identity markers. This is because much of the period’s economic growth occurred near London. Because SE culture was already widespread, migration toward this region amplified its popularity due to coordination dynamics. Economic development favors broader adoption of identities shared by many.

Our counterfactuals also reveal that industrialization's effect on local cultural loss is location-dependent. In central regions, high rates of in- and out-migration facilitate cultural turnover. In contrast, industrialization in peripheral areas attracts fewer migrants and retains more of the local population—thereby helping to preserve regional identities. This worked to counteract broader developments during this period. Our model shows that falling migration costs—a distinct component of modernization—exerted broad pressure on the periphery to adopt the SE culture. Where industrialization occurred locally, however, it acted as a counterforce—slowing or even stalling peripheral cultural convergence.

Our framework and findings help illuminate why some industrializing peripheries retained distinctive cultural identities. One such case, described in work by Elliott Green, is the Basque Country, where the development of ore production and steel manufacturing drew migrants from the surrounding region into Bilbao—diverting movement away from Madrid and reinforcing local cultural cohesion. Similar dynamics played out in Catalonia and Scotland.

More broadly, our study clarifies how bottom-up forces shape the cultural map, even in the absence of strong state-led nation-building. While our model abstracts from widely studied mechanisms like education, it still explains a substantial share of the identity shifts observed in 19th-century Britain. The interplay between uneven economic development and growing mobility remains central in today's world—particularly across the Global South. A framework like ours offers a foundation for understanding how cultural identities evolve in response.

Creative workers tend to co-locate geographically: computer science researchers cluster in Silicon Valley, literary writers in New York, and country musicians in Nashville. Agglomeration facilitates knowledge spillovers, enables faster and better matching between firms and workers, and generates economies of scale in the provision of shared inputs such as computing clusters, publishers, and recording studios. Through these agglomeration economies, the spatial concentration of creative workers enhances productivity and innovation.

Despite their central role in modern urban economics, we know relatively little about agglomeration economies in historical contexts, particularly before the 18th century. It is evident that the arts and sciences experienced periods of efflorescence in renowned cities, such as Florence during the Renaissance. However, systematic data from these earlier periods remain scarce.

In an ongoing research agenda, in collaboration with Gary Cox, we begin to address this gap by assembling the largest panel database of published authors from the Scientific Revolution and the Enlightenment (c. 1500–1800). Drawing on large bibliometric databases and national biographical dictionaries, we study thousands of book authors. Our panel tracks authors throughout their entire careers, recording—for each year of their lives—the cities in which they lived, their collaborators, the learned societies and educational institutions they joined, and the books they published.

Agglomeration and Creativity in the British Isles

Our first study, which appeared in Explorations in Economic History this year, focuses on London’s knowledge economy. We examine early modern London because it featured all three main sources of agglomeration economies: knowledge spillovers, matching potential collaborators, and sharing indivisible inputs.

London was an epicenter of creative work. For anyone hoping to meet European scholars visiting Britain, the chances were better in London than anywhere else. The city also boasted a plethora of social clubs, cafés, and learned societies where scholars and intellectuals gathered to exchange ideas. These venues (and the relatively weak regulatory power of metropolitan guilds) made it easier to find interlocutors and collaborators. For example, the engineer John Smeaton began attending meetings at the Royal Society immediately after moving to London.

Productive interactions were not limited to those within one’s own field. According to a recent book, for example, William Shakespeare’s neighbors in the parish of St. Helen’s, London, may have influenced his plays. One such neighbor, Edward Jorden, was among the first English doctors to specialize in female health and was called as an expert witness in high-profile witch trials. A historian has conjectured, based on contextual evidence and the content of Shakespeare’s and Jorden’s writings, that their relationship might have shaped the portrayal of the “three witches” and the psychological depth of Lady Macbeth in Macbeth.

In addition to knowledge spillovers and advantages in matching with potential collaborators, London also offered cutting-edge public goods valuable to writers and researchers. For example, the city was home to England’s first professorship in mathematics (the Gresham Chair in Geometry, established in 1596), its first learned society (the Royal Society, founded in 1660), and its first astronomical observatory (the Royal Observatory in Greenwich, established in 1675). Moreover, authors could sell their works through London’s many bookstores, and playwrights could stage their plays in London’s many theaters.

To quantitatively assess whether residing in London benefited authors, we drew on our panel database of British authors from the 1470s to 1800. In this phase of our research, we focused exclusively on prolific authors—those who published at least five works listed in the English Short Title Catalogue. For these individuals, we reconstructed full residential histories using biographies from the Oxford Dictionary of National Biography. For illustrative purposes, Table 1 presents the top five authors by field, ranked by the availability of their works in modern library holdings.

We follow the standard econometric approach in urban economics and regress an author’s productivity on residence in London, controlling for year and author fixed effects. Intuitively, we measure how much more (or less) productive authors became upon moving to (or from) London, compared to other authors who didn’t move that year. The event-study plot in Figure 1 shows that authors were typically not becoming more productive before moving to London. Once they moved there, however, they experienced a sharp and persistent productivity boost that lasted for at least ten years.

The “London premium” we estimate represents a 37% boost relative to the average annual productivity in our sample. Our estimate of the London premium persists when we control in our regressions for age and age-squared, time-varying wealth shocks, and when we exclude authors from the sample after they presumably retired or before they graduated from university. Interestingly, London’s premium did not always exist¾it emerged in the second half of the sixteenth century.

Why did living in London confer such a premium? We provide evidence that authors residing in London benefited from knowledge spillovers and enhanced collaboration opportunities. As shown in the event-study analyses in Figure 2, authors who moved to London became more likely to form close intellectual collaborations (left plot), particularly in the first few years after their move, and to join the Royal Society (right plot).

London also afforded advantages for publishing books, once they were written, because the London’s Stationers Company had monopoly rights over printing between 1577 and 1711. Yet non-London authors in our sample published two-thirds of their books in London, and the city’s premium persisted after 1711.

A Cross-Regional Analysis of Agglomeration Economies

Another part of our project adopts a cross-regional approach. Using harmonized data from different regions, we examine changes in the knowledge economies of Britain and the Netherlands (in the northwest) and Spain and Italy (in the south) from 1500 to 1700. The data we have explored so far suggests that urban knowledge economies in the south began to diverge from those in the northwest in the seventeenth century. In the south: (i) intellectual interactions became less predictive of published output; (ii) clustering of authors in creative cities diminished; and (iii) urban productivity premiums—authors' boost in output when they moved to leading cities—declined.

These changes suggest a significant transformation in intellectual labor markets that has not yet been explored in the historical literature on the “Little Divergence” (the take-off of living standards in the European northwest relative to the south before the Industrial Revolution). We believe the divergence we observe may have political origins, largely stemming from transformations that began with the Protestant Reformation and the Catholic Counter-Reformation. In Protestant regions, ecclesiastical confiscations triggered a process of human capital reallocation, flooding the secular labor market with educated individuals (see, e.g., here and here) who had greater flexibility in choosing where to work than centrally appointed clerics. In contrast, in Catholic regions, the repressive response to the Reformation created significant barriers to the intellectual interactions that are central to agglomeration economies.

The Spanish Inquisition and Intellectual Isolation

We have explored political barriers to intellectual activity in a third project, which examines the consequences of the Spanish Inquisition’s enforcement of Catholic orthodoxy.

Founded in 1478 and initially focused on persecuting suspected Jews, the Spanish Inquisition shifted its efforts toward Protestantism after the onset of the European Reformations, and particularly during the Council of Trent (1545–1563). The discovery of Protestant networks in two Spanish cities in 1557-58 prompted a dramatic and highly publicized increase in inquisitorial activity. Authorities responded to this shocking discovery by banning travel to foreign universities, publishing an index of forbidden books, and declaring that possession or reading of these books would be punishable by death. Additionally, fourteen people were burned at the stake in Valladolid during a massive auto de fe designed to awe; persecution of foreigners suspected of being Protestant sharply increased; the highest Catholic authority in the Spanish Empire, the Archbishop of Toledo, was arrested in part for his interactions with suspected Lutherans; and inquisitorial activity reached peak levels.

Given that the contours of heresy were blurry and continuously redefined, and that even interaction with suspected heretics was policed by the inquisitors, we argue that scholars should have reacted to inquisitorial changes by limiting their contacts and by exiting certain fields and institutions. This should have hurt STEM output the most because interacting with others working on similar problems is essential to progress in the sciences.

Figure 3 summarizes our main findings. The left panel shows that the share of STEM book publications in Spain declined after the 1560s relative to Protestant Europe. The right panel shows the proportion of Spanish authors whose biographies mentioned an instance of intellectual interaction each year. After the transformation of inquisitorial persecution in 1558-9, we detect a statistically significant reversal in the previously upward trend in interactions; and a pronounced immediate drop in the propensity to interact with other intellectuals. We also detect a migration of authors out of universities (secular educational institutions) into convents and religious colleges.

Our findings contradict a revisionist understanding of the Spanish Inquisition (e.g., this), which argues that its targeting and punishment of scientists has been grossly exaggerated; that it lacked the capacity to enforce book bans; and that it likely had little effect, given that Spain’s Golden Age (1492-1657) began after the Inquisition was founded (1478) and co-existed with it for over 150 years. Instead, we argue that the Inquisition’s direct targeting of specific scholars, theories, and books was less important than the chilling effects that such efforts produced: inducing scholars to reduce their interactions with anyone the Inquisition might scrutinize; and prompting various forms of self-censorship.

Too Long, Didn’t Read?

Creative workers benefit from free labor markets that allow for mobility across locations and jobs, the opportunity and freedom to interact with others (both within and across fields), and shared creativity-complementing amenities. This seems to also have been the case in early modern times. Even in 1600, creativity thrived where people came together — but especially in London’s learned societies, not the cloistered halls of Spanish convents.

Valentín Figueroa is currently an assistant professor at the Pontificia Universidad Católica de Chile. In 2025, he will join the Massachusetts Institute of Technology as an assistant professor. He holds a Ph.D. in Political Science from Stanford University. His website is www.valentinfigueroa.com

“Frontiers were, indeed, the razor’s edge upon which hung suspended the modern issues of war and peace, life and death”, Lord Curzon, “Protectorate and Hinterland”, Romanes Lecture, University of Oxford (1907).

“To speak of frontier governmentality in the modern world, then, is to speak of a long history of violence.'' Benjamin Hopkins, Ruling the Savage Periphery: Frontier Governance and the Making of the Modern State. Harvard University Press (2020).

For centuries, borderlands have been cauldrons of rebellion, resistance, and violence against the state. Recent data confirms this: a 2024 OECD Report found that peripheral regions of modern states experience significantly more conflict than central territories. Since 2010 the intensity of such violence has also increased in border regions, which have proven to be fertile grounds for armed insurgencies. Why do frontiers tend to experience greater violence against the state than heartland territories? In our research, we offer a deep explanation rooted in the history of colonial rule. Imperial powers often ruled their frontier territories, which were typically liminal spaces on the edges of empires, differently from the core colonial regions. Such “rule of difference” typically rested on distinct administrative and legal practices. Historian Benjamin Hopkins describes this distinct frontier rule as “frontier governmentality”—a laissez-faire institutional arrangement where the state’s presence was relatively thin, authority was delegated to local elites, and frontier residents were cut off from formal institutions of conflict management (e.g. courts, police, and electoral politics). Such frontier buffer zones effectively operated under conditions of partial sovereignty whereby colonial rulers shared with local elites the state’s power over coercion and social control. It was thus a highly personalized form of rule where the state delegated even greater power to local elites than was the case under indirect rule. These arrangements, rooted in colonial cost-benefit calculations (e.g., high governance costs or external threats), became a global template, applied from British India’s North-West Frontier to Kenya’s northern borderlands. Crucially, this institutional legacy endured post-independence, creating uneven territorial governance within former colonies.

While historians, anthropologists, and political scientists have examined the enduring legacy of frontier rule, its role in shaping contemporary conflict has not received any empirical attention. Investigating this relationship in a rare empirical context, we argue that while frontier rule can sustain social order over long periods, it is especially vulnerable to violence when disrupted by external shocks. As a distinct mode of governance, it relied on minimal state presence and elite mediation—trading institutional depth for fragile stability. This limited resilience stems from several factors. Frontier populations often maintained only a tenuous connection to the state due to its shallow reach, shaping state–society relations in profound ways. Unlike residents of non-frontier regions, those in frontier areas typically lacked access to formal institutions such as courts or electoral systems, instead depending on traditional dispute resolution led by local chiefs and elders. In contexts where social order rests almost entirely on elite mediation, any threat to those elites risks creating an institutional vacuum—one that can quickly become fertile ground for violence. In the absence of institutionalized channels for negotiation and competition, disaffected groups are more likely to challenge the state’s legitimacy through sovereignty-contesting violence.

The setting: British India’s Northwestern Frontier

Our research draws on one of original and classic cases of frontier rule established in British India’s North-West Frontier Province, bordering Afghanistan. In 1901, the British divided the province into settled districts and frontier areas, both inhabited by Pashtuns but governed under sharply divergent institutional regimes. For over a century, the frontier tracts were ruled—first by the British Empire, then by Pakistan—through local elites rather than courts, police, or bureaucracies. While settled areas had formal administrative structures, legal protections, and limited political representation, frontier regions operated under minimal state presence, relying on tribal elites, informal jirgas (Councils of Tribal Elders), and paramilitary forces. This stark institutional discontinuity was aptly captured by Sir Olaf Caroe, a colonial-era governor, who observed that “the line of administration stopped like a tide almost at the first contour” of the frontier. This enduring “rule of difference” legally disenfranchised frontier populations and excluded them from judicial and electoral systems. It persisted long after independence, with frontier areas remaining legally and politically marginalized well into the 21st century, despite formal reforms in 2018 that have yet to produce meaningful change on the ground.

Although frontier rule was widely applied across colonial contexts in the late 19th and early 20th centuries, its long-term effects on conflict are difficult to assess because frontier boundaries were often non-random, shaped by natural geography (e.g. waterbodies, rivers, or mountains), pre-colonial borders, or mapped with scientific precision. Pakistan’s North-West Frontier offers a rare exception. While broader strategic considerations may have motivated the British to establish a frontier buffer zone, we argue that the actual placement of its border was determined in a haphazard and arbitrary manner. Drawing on a rich body of historical evidence, we show that the British employed a narrative-based, non-scientific cartographic regime in mapping these frontiers. Unlike the rest of British India, which relied on precise trigonometric surveys, the frontier regions were subjected to a hurried and improvised route-mapping, relied on inconsistent local knowledge, and were shaped by the discretion and idiosyncrasies of surveyors, guides, and administrators (see Figure 1). This imprecise method reflected colonial perceptions of the frontier as a mysterious, adventurous, and unknowable buffer zone rather than a space to be accurately mapped and systematically governed. Historian Thomas Simpson aptly characterizes this process as “cartographic anarchy”, underscoring the “chaos and heterogeneity of survey work on the ground in British India”. Crucially, the frontier rule border did not align with pre-colonial boundaries or natural geographic features. While some broad differences exist between frontier and settled areas, the internal terrain of the frontier varied significantly—undermining any simple ecological divide, such as the hills-versus-plains distinction proposed by James Scott.

Figure 1: This figure presents the trigonometrical survey map of British India from 1881. The map highlights a stark contrast: the northwestern frontier is represented with elongated lines, indicating imprecise mapping, while the rest of British India is mapped precisely using triangles. Source: General Report on the Operations of the Great Trigonometrical Survey of India, 1881.

Our empirical analysis leverages this arbitrarily defined historical border in a spatial regression discontinuity design to examine whether historical exposure to frontier rule predicts contemporary conflict in closely situated regions. In this regard, we utilize a geo-coded data on conflict against the state, measured at a highly granular level (i.e., 10 x 10km grid-cells) during the period 1970-2018. Figure 2 shows the area of study and visually represents the spatial distribution of conflict incidents.

Figure 2: Panel A shows the area of study within which our sample is restricted. It encompasses all those areas that lie within a 50-km buffer zone around the FR boundary. Panel B shows the spread of conflict incidents against the state from 1970 to 2018. Each black dot represents an attack against the state (defined here as military personnel or installations) and the red line denotes the historical FR border (as of 1901) separating frontier areas (left side of the border) from settled districts (right side of the border).

Frontier rule and conflict

Our results show that, on average, between 1970 and 2018 areas that fell just inside the historical frontier border witnessed a significantly higher incidence of conflict against the state than areas just outside the frontier border (see Figure 3). The effects are substantial: residents just inside the border demarcating frontier rule were 57 per cent more exposed to conflict than those just outside the border. To ensure that this difference is not due to other underlying factors, we test for and find no major differences across the border in geography, climate, population, or social structure (e.g. ethnicity and religion). We also show that areas on both sides of the FR border display no significant differences in relevant historical dimensions, such as pre-colonial conflict and population density.

Figure 3: Binned scatterplot (16 bins of size 5km each) of the unconditional relationship between conflict against the state and distance to the FR border. The y-axis reports the natural log of 1 plus the incidents of conflict against the state. The x-axis reports the distance (in km) from the FR border for areas under FR and non-FR. The border itself is at km 0 with positive values indicating km inside the FR territory.

To unpack our results and probe their substantive meaning, we explore the temporal dimension and show that this differential conflict incidence between frontier and non-frontier regions is a feature of the post-9/11 period (see Figure 4). Specifically, the impact of frontier rule on conflict emerged only after the 2001 U.S. invasion of Afghanistan, which was widely unpopular in Pakistan’s Northwestern regions where it was seen as an unjustified attack against ethnic Pashtun brethren across the Afghan border. The year 2001 marked a watershed moment when, under external pressure, Pakistan’s military ruler reversed the country’s decades old Afghan policy, joined the U.S.-led War on Terror, and extended the military’s presence into frontier areas. While initially the Pakistani military only increased its physical presence on the border to prevent Afghan fighters spilling over into Pakistan, it was nevertheless seen as an alien force doing the bidding of a foreign power.

Figure 4: This figure plots the point estimates and confidence intervals of the frontier rule (FR) dummy interacted with year fixed effects for an event study specification that also includes district and year fixed effects. The coefficient for the base year (1970) was set to zero and is not shown in the figure. The analysis is conducted at the sub-district (tehsil) level. The sample includes all sub-districts of Pakistan. The coefficient on the FR interaction dummy progressively increases from 2001 onwards until it becomes statistically significant in 2008 and reaches its peak in 2012.

While the post-9/11 grievance against the state was present in both frontier and non-frontier regions, it translated into heightened conflict in frontier areas that had historically suffered from an institutional void. A key driver of this escalation was the targeted elimination of tribal leaders—critical intermediaries between the state and local communities—whose removal, in the absence of formal conflict-resolution mechanisms, deepened instability and violence. To lend empirical credence to this line of reasoning, we offer two types of evidence. Firstly, we use data from a nationally representative survey to shed light on the three dimensions that make frontier rule vulnerable: absence of formal institutions of conflict management, greater dependence on elite intermediation, and low trust in state institutions among frontier residents. Secondly, we empirically demonstrate how the strategic elimination of tribal elites unravelled social order in frontier areas (see Figure 5).

Figure 5: This figure plots the coefficients and confidence intervals for the estimated relationship between the first-ever attack on a tribal elder and the overall level of conflict in the same grid cell. The dependent variable is ln (1 + all conflict incidents). Estimates are from a regression model that allows for effects before, during, and after the first-ever attack on a tribal elder and includes district and year-fixed effects.

Recognizing the complexity of insurgency-based violence, we entertain multiple competing explanations behind the post-9/11 rise in violence against the state in the frontier rule areas. Foremost among these is the possibility of a conflict spillover from Afghanistan. We show that an overwhelming majority of attacks against the state were carried out by local outfits rather than Afghan-based militants. Next, using a variety of empirical approaches, we also rule out the potential role of U.S. drone attacks and Pakistani military operations in the area (which also constituted an important income shock for local populations). We show that both drone attacks and military operations against terrorist outfits were an endogenous military response to the rise of violence, which pre-dated the original post-9/11 uptick in violence. Finally, we address concerns that persistent under-provision of public infrastructure in geographic peripheries like Pakistan’s north-west frontier may be driving anti-state violence. We find no statistically significant differences in infrastructure, whether modern (roads, waterways, health centres) or historical (colonial railways, Mughal roads, pre-colonial Islamic trade routes), between frontier and non-frontier areas.

Implications of our study

We conclude this blog by highlighting the broader relevance of our study for three important areas: (a) understanding violence in other frontier regions; (b) contribution to relevant academic literatures, and (c) historical political economy.

Our findings speak to a much broader story than just Pakistan’s frontier regions. Similar patterns of “frontier rule” have existed in places like Northern Kenya, Northern Nigeria, and Iraq’s Basra region—areas where colonial powers created special systems of governance and carved out hybrid zones where state authority was always partial, always negotiated. These arrangements, built during the 19th and early 20th centuries, often kept the peace—until droughts, wars, or political shocks exposed their fragility. Today, as climate change and geopolitical rivalries strain borderlands worldwide, understanding these institutional legacies isn’t just an academic curiosity but also relevant for policymakers. It’s a warning: in regions where the state never fully established control and where relationships between the government and local communities remain fragile, even isolated shocks can turn peripheries into powder kegs. Crucially, resolving conflict in such frontier regions isn't just a matter of military action. Instead, enduring peace requires a more essential and difficult task: confronting deep-rooted institutional gaps and rebuilding trust where, for generations, the state has been absent or alien.

Our research has important implications for the conflict literature, which emphasizes the role of institutions but has yet to fully examine the impact of specific legal and political structures. As a comprehensive review by Blattman and Miguel argues, key institutional features are ‘yet to be carefully defined and measured’ in the conflict literature. Furthermore, as they note, the impact of institutions on conflict can be conditional on other factors. To this end, our research underscores how the interaction between historically embedded governance systems in frontier areas can interact with geo-political shocks to fuel violence against the state. Our work also directly contributes to a growingly niche literature on frontiers and their influence on development outcomes, including long-run inequality, economic geography, public infrastructure, and culture and politics.

Our work also connects with emerging conceptual and methodological debates in historical political economy (HPE). Our work moves beyond persistence studies, which simply trace the long-run impact of historical legacies, to offer more nuanced evidence on “why” and “how” history matters for contemporary violence. In this regard, our work aligns with the emerging focus in HPE on historical contingency and “time-varying persistence”, demonstrating how historical legacies can remain latent until activated by specific shocks—such as, in our case, the geopolitical shock of 9/11. We also underscore the importance of paying attention to timing and sequence, which can be critical for making causal claims in the context of insurgency-based violence where several mutually reinforcing factors are triggered in a sequence drawn out over time. Paul Pierson’s advice is particularly relevant here; he argues that “long-term outcomes of interest depend on the relative timing of important processes [. . .] A variable’s impact cannot be predicted without an appreciation for when it appears within a sequence unfolding over time.” One of our final learnings while working on this project was the importance of qualitative historical evidence in validating natural experiments that rely on the assumption of “as-if-random” borders (e.g., in spatial regression discontinuity designs). In this context, a deep dive in history is not a luxury but a necessity, especially in the light of recent work that has overturned the prior academic consensus on randomly-defined colonial borders in Africa.

Authors:

Adeel Malik is associate professor of development economics at the University of Oxford and Globe Fellow in Economies of Muslim Societies at the Oxford Centre for Islamic Studies. His research focuses on the political economy of development in Muslim societies, particularly in the Middle East and North Africa. His recent work on Pakistan has probed the deep and enduring legacies of colonial rule. He has published in esteemed academic journals, including Journal of the European Economic Association, Journal of Development Economics, Journal of Historical Political Economy, the European Journal of Political Economy, and the Journal of Democracy. Additionally, his research has been featured in high-profile journalistic outlets such as the CNN, Financial Times, the New York Times, the Washington Post, Project Syndicate, Foreign Policy, and Foreign Affairs.

Rinchan Ali Mirza is a Lecturer (Assistant Professor) in Economics at the University of Kent, having previously served as a post-doctoral researcher at the University of Namur’s Center of Research in the Economics of Development from 2016 to 2019. He holds a DPhil and an MPhil in Economics from the University of Oxford, as well as a BSc in Mathematics and Management from King’s College London. His research spans economic history, political economy, the economics of religion, development economics, and applied microeconomics. Dr. Mirza is a faculty fellow at the Association for Analytic Learning about Islam and Muslim Societies (AALIMS), a research fellow at the Centre for Economic Research in Pakistan (CERP), and a founding member of the Development Economics Research Centre at Kent (DeReCK), reflecting his commitment to rigorous, policy-relevant research on development and the Muslim world.

Faiz Ur Rehman is an Associate Professor of Economics in the Department of Economics at IBA. He holds a joint Ph.D. degree in Law & Economics from the Universities of Bologna, Hamburg, and Erasmus Rotterdam, acquired under the Erasmus Mundus Doctorate Fellowship. He joined IBA in 2021, and prior to that, he was associated with Quaid-i-Azam University Islamabad. His areas of interest encompass the economics of conflict, political economy, and applied economics. Presently, his research focuses on projects including Frontier Governmentality, Frontier Rule and Long-Term Development, Conflict and Early Human Capital, as well as Firms' Public Credit and Default Rate. His instructional portfolio at IBA covers Macroeconomics, Applied Economics, and Institutions and Development.

The profound political, economic, and social changes that followed have inspired thousands upon thousands of studies. We understand much more about the demographic impact of conquest and colonization on the American indigenous populations. We have learned much about the forced migration of millions of Africans.[1]

But, perhaps surprisingly, we lack knowledge about a key group in this process: the European migrants. This is where we come in. Funded by the National Science Foundation and the Economic Growth Center at Yale University, we are working on a project to shed light on the European migrants to Spanish America. Starting with migrant zero, Christopher Columbus, we have gathered microdata on the hundreds of thousands of Europeans that ventured to the “New World.”

We want to share our intellectual adventure and some insights we have learned so far.

Inspiration and perspiration

Many scholars (and others) have blamed Latin America’s disappointing economic performance since independence to the traumatic colonial origins (Engerman and Sokoloff 1997, Acemoglu et al. 2001). We talk about “bad” colonial institutions, unfavorable factor endowments, and risky production profiles. It is of course undeniable that the Spanish colonization of the Americas was shaped by the available resources. But was the impact of the interaction between the colonizers, the colonized, and the enslaved really so clear cut? While we have an extensive scholarship on selected locations within the Spanish Empire, we lack a general knowledge about the origins and characteristics of the colonizers. Our lack of knowledge about the colonial migration contrasts with our wealth of data regarding the later age of mass migration (1860-1920). This vital episode in history lacks a systematic analysis of the migrant flows that changed the world.

Our project is mainly based on the transcriptions of thousands upon thousands of unpublished records housed at the Archivo General de Indias (AGI) in Sevilla, Spain. We have assembled the first database of individual migrant records with a rich set of characteristics including sex, occupation, origin, destination, and year of departure. For the early period, 1492 to 1540, we have transcribed all the individual records from Boyd-Bowman (1964), Boyd-Bowman (1968), and Archivo General de Indias (1940). Due to the nature of the record-keeping practices in the early period and unfortunate fires at the AGI, we also consulted secondary sources to include the four Columbian voyages (1492, 1493, 1498, and 1502), the Ovando expedition, and other esoteric sources like contracts of carriage housed at the archive of notary records in Sevilla. Over 90% of all our observations come from primary sources: licenses, ship manifests, and contracts (see Figure 1).

Other scholars have tackled this topic. Boyd-Bowman (1964, 1968), using secondary sources, put together thousands of records of voluntary migrants throughout the first century of conquest and colonization. His magnum opus, Indice Geobiográfico de Cuarenta Mil Pobladores Españoles de América en el Siglo XVI, is the first attempt to quantify the migration flows to Spanish America. Marques Macías (1995) resumes this study for the late colonial period (1765-1824). We also have rough back-of-the-envelope estimates by Engerman (1999), Eltis (1989), and Morner (1975). All these estimates, however, fall short of a detailed picture of migration flows that lasted over three centuries of Spanish rule (1492-1820s).

Translating sources

Our project heavily relies on the state apparatus created to regulate migration to the Americas. The Spanish Crown highly regulated migration to have the “right” kind of people settled in the newly-acquired territories. This translated to Catholic, qualified, and from the unified kingdoms of Castilla and Aragon.

In order to travel, a prospective migrant needed to apply for a “visa.” This application attested his or her purity of blood (at two generations Catholic), good character (no outstanding debts or criminal history), and documentation on origin (marital status, noble status, and other family history). These “visa” applications are one of our main sources for our project, known as licencias, licenses.

These licenses provide a wealth of information on individual migrants and their companions. Let’s explore one of them. In 1587, Alonso Rodríguez de Vera (see Figure 2) applied to go to Peru to join his brother on his hacienda. His application for travel includes his wife, children, and a female servant. He also took the opportunity to request lands at destination and to request authorization to bring along weapons (two swords, two daggers, and an arquebus).

Once the “visa” was approved, the migrant had to buy a “ticket.” Housed in a different archive, also in Sevilla, we have found and transcribed thousands of contracts of carriage signed between migrants and shipowners. They provide very useful information. In the case of Isabel Díaz, the contract of carriage between the Sevillian passenger and Pedro de Umbría –first officer of the ship Santa María de Guadalupe—stipulated payment of 15 gold pesos for travel to the port of Santo Domingo (in the present-day Dominican Republic). From this document we also learned that Isabel was illiterate as Pedro de Oviedo signed it on her behalf (see Figure 3).

Finally, the migrant had to board the ship. That is another of our sources: the ship manifests –known as asientos. These documents have similar information as the licenses but in an abridged form and listed by ship departed from Sevilla. All these sources allow us to build a fairly detailed picture of voluntary migrants traveling to Spanish America.

Trends

This labor of love allows us to assess the size of the migrant flows from Spain to the Americas for the entire colonial period. We can attest that over 250 thousand free Europeans left their homeland to try their luck in the New World (see Figure 4). The first four decades we observe migrants trickling in, moving to an ever-expanding menu of destinations. The flows accelerate for a century to then increase but at a slower pace for the remainder of colonial rule.

Regarding origin, as it is well-known, the initial conquest was mostly a Castilian phenomenon. Although, there were no restrictions on travel from other Spanish kingdoms, network effects were strong, and the early migrants were mostly Castilian. Migrants from other regions were much more likely to be merchants. Early migrants were mostly males, and mixed with the locals throughout the 16^th century. This pattern changed, starting in the mid 17^th century, the typical migrant travelled with his entire household increasing the female migrant share to 50%. We see a similar pattern for the clergy, with very few friars crossing the pond after 1650.

As for destinations, we discovered that following the footsteps of the conquistadors, the first twenty thousand migrants moved first to the Caribbean, then to Tierra Firme (modern-day Central America + Colombia + Venezuela) to then arrive to what would become the pillars of the Spanish empire: Mexico and Peru. But that pattern did not last. Over the entire period, the bulk of migrants chose as their main destination Mexico and Peru: almost 60% of all migrants declared these two destinations (see Figure 5).

Our estimates are far below the existing upper bound estimate of 746,000 people by the late colonial period. Granted, Engerman’s figures are based on Eltis’s own estimates obtained from population data of the Americas and Morner’s own ship size estimations. We cannot claim that our detailed research will capture every single European that ventured to Spanish America. We are, however, confident that we are offering a trustworthy lower bound estimate. To assess the robustness of our estimates, we replicated Morner’s numbers (based on Chaunu’s accounting). We refined these estimates by removing selected ships categories that were unlikely to carry passengers: slave ships and those carrying mercury from the mines in Almadén in Spain to Mexico. In addition, we adjusted ship sizes using recent studies which found that while the nominal tonelada castellana was getting bigger, the actual ship tonnage remained the same. This resulted in tonnage nominal inflation of 60% by 1620 and coincides exactly with Morner’s inflated estimates of likely migrants.

Searching for answers

Our project has opened more and more questions as we moved along. We have a pipeline of projects with different co-authors to exploit this rich dataset.

We have started with the early period, the first waves of migration. We look at the first movers and settlers, starting with Christopher Columbus and ending with a decade after the conquest of Peru. In “Gambling for America,” joint with Yannay Spitzer and Ariell Zimran, we characterize the early migration flows to find that a clear pattern emerges. Networks of origin develop quickly with one-third of all migrants coming from two provinces –Sevilla and Badajoz. These migrants did not have much knowledge about the destinations per se. Rather, they headed out to new destinations almost randomly as the frontier opened up new areas for settlement.

With Desiree Desierto, we are looking at labor coercion practices in Spanish America. As the conditions on the ground varied considerably throughout the empire, different equilibria emerged leading to a spectrum of extraction. With Noel Maurer and Miguel Angel Navarro, we are starting a paper studying forced migration to the New World. Using a companion dataset on enslaved migrants, we are tracking these flows throughout colonial times. We are also working on a big-picture article taking into account different inflows of migrants, enslaved and voluntary, and the demographic trends that emerged after the Conquest. In addition, the imperial expansion also had effects in the motherland, in terms of population movements, we are interested in analyzing how emigration to the Americas affected demographic dynamics within Spain.

Slowly but surely, we are tackling important topics about migration in colonial times. We are populating our project website as we go along, please check it out: Bridging the Atlantic: Migrations & their Legacies.

So stay tuned! And if you have a brilliant idea for collaboration, please do reach out.

References

Acemoglu, Daron, Simon Johnson, and James A. Robinson. “The colonial origins of comparative development: An empirical investigation.” American Economic Review 91, no. 5 (2001): 1369-1401.

Boyd-Bowman, P. (1964). Indice Geobiográfico de Cuarenta Mil Pobladores Españoles de América en el Siglo XVI, Tomo I, 1493-1519. Bogotá: Instituto Caro y Cuervo.

Boyd-Bowman, P. (1968). Indice Geobiográfico de Cuarenta Mil Pobladores Españoles de América en el Siglo XVI, Tomo II, 1520-1539. Mexico: Editorial Jus. Mexico.

Eltis, David. Coerced and free migration: Global perspectives. Stanford University Press, 2002.

Engerman, Stanley, and Kenneth Sokoloff. “Factor Endowments, Institutions and Differential Paths of Growth among the New World Economies,” in Stephen Haber, ed., How Latin America Fell Behind, Stanford: Stanford University Press." (1997).

Mörner, M. (1975). La emigración española al nuevo mundo antes de 1810. un informe del estado de la investigación. Anuario de estudios americanos 32, 43 131.

[1] Thanks are particularly due to the impressive scholarly collaboration in Slave Voyages.